Word Order in Sanskrit and Universal Grammar by J.F.Staal
Word Order in Sanskrit and Universal Grammar by J.F.Staal
UNIVERSAL GRAMMAR
)
.
FOUNDATIONS OF LANGUAGE
SUPPLEMENTARY SERIES
Editors
MORRIS HALLE, MJT
PETER HAR TMANN, MiinsterjW.
K. KUNJUNNI RAJA, Madras
BENSON MATES, Univ. of California
J. F. STAAL, Amsterdam
PIETER A. VERBURG, Groningen
JOHN W. M. VERHAAR (Secretary), Manila
Ateneo de M ani/a University
VOLUME 5
J. F. STAAL
WORD ORDER IN
SANSKRIT AND UNIVERSAL
GRAMMAR.
D. RE !DEL PUBLISHING. COMPANY /DORDRECHT-HOLLAND
1967
All rights reserved
No part of this book may be reproduced in any form, by print, photoprint,
microfilm, or any other means without permission from the publisher
Printed in The Netherlands by D. Reidel, Dordrecht
-
- -- ------
To Saraswathy
'
)
niidasya kramajanmatvdn na pilrvo na paraS ea sal:z
akrama/:z kramariipe1Ja bhedaviin iva jiiyate
'Though sound is prodaced in a fixed order
Speech itself has no earlierand later.
Itself without order, it is produced
As if divided by the appearance of order.'
Bhartrhari, Vakyapadiya 1.48
1st die Reihenfolge der Grundbestimmungen, nach deren Modell ein jeder Satz
modelt wird, nicht ebenso willki.irlich, wie die der jeweils zu interpretierenden Satz? Hier
ist zu antworten, dass das Vorangehen eines bestimmten Grundmomentes und das Folgen
eines anderen nie aus dem Prinzip der Reihenfolge ableitbar ist .... So kann
etwa im Kreise unserer Sprachen als ers tes Grundmittel erkHirenden Verstehens das
Subjekt augewendet wen.len, insofern sich das Subjekt tatsachlich orter am Anfang des
konkreten Satzes be:findet. Das vom Subjekt geforderte Zweite, das Priidikat, aber steht
in unseren Sprachen rneistens hintenan, und hiermit zeigt sich die lnnerlichkeit des
zips der Anordnung der grammatischen Interpretation: ihr fester Weg ist in seiner
keit eben ein anderer als der empirisch wandelbare und von Fall zu Fall sich wandelnde
Weg im Aufbau eines Satzes.
H. J. Pos, tlber den tJ.ufbau der grammatischen
Interpretation, p. 303
PREFACE
This monograph owes its existence to certain puzzles in universal grammar
and the theory of language which led the author to an investigation of
word order in Sanskrit and its possible analyses and descriptions. Not
unexpectedly, the raw material was found to be too vast for a first-hand
treatment even to be attempted. Rather surprisingly, however, its inter-
pretations by Indian and Western theorists and grammarians turned out to
be so greatly at variance, that an analysis of these interpretations seemed
rewarding. Accordingly, theoretical issues within the framework of generative
grammar had to be faced anew, and alternative solutions suggested them-
selves. In this connexion the Sanskrit grammarians proved not only in-
spiring but positively helpful.
This book may invite the accusation that it wilfully mixes disciplines.
There were alternatives: one could try to write a history of the subject; or
construct a merely formal edifice, leaving it to others to test its adequacy;
or else one could make the notorious attempt to stick to the facts, which
is not only unilluminating but also bound to fail. Any such self-imposed
restrictions seemed to conflict with the original intent. And so it was decided
not only to make available the results of the investigation into Sanskrit
word order; but also to introduce a theory of universal grammar to account
. for these and other results.
I should like to thank Professors C. E. Bazell, Noam Chomsky, Morris
Halle, A. L. V os, Dr. A. Kraak and Mr. P. G. E. Wesly for their suggestions,
criticism and opposition. It will be obvious where I am indebted to them, and
especially to Chomsky. Bibliographical information has been supplied by
Professors S. A. Boaebakker, Barron Brainerd, A. L. Vos and by Dr.
D. F. W. van Lennep, to whom I am also very grateful.
1966, Brighton-Amsterdam J. F. STAAL
IX
TABLE OF CONTENTS
I. INTRODUCTION 1
Il. INDIAN THEORISTS ON WORD ORDER IN SANSKRIT 16
A. f!.kpriitisiikhya, Nirukta, Brhaddevatii 21
B. Pa1.1ini and Pataiijali 26
C. Jaimini, Sahara, and Kumarila Bhatta 45
Ill. WESTERN SANSKRITlSTS ON WORD ORDER IN SANSKRIT 51
IV. SUMMARY AND CONCLUSIONS 60
APPENDIX: SOME SEMANTIC RELATIONS 81
BIBLIOGRAPHY 91
A. Texts and Translations 91
B. Other Works 91
INDEXES 95
A. Names 95
B. Technical Terms: Sanskrit 96
C. Technical Terms: English 97
D. Languages 98
XI
CHAPTER I
INTRODUCTION
Vaman Sivaram Apte was among the most illustrious of those Sanskrit
scholars who were at the same time Sanskrit speakers. His 'Student's
Guide to Sanskrit Composition', first published in 1881, was re-edited and
reprinted at least twenty-five times and continues to be used in high schools
and colleges in India. In 399-401 (from the second edition ofl885 onwards)
of this remarkable treatise on Sanskrit syntax the following statements
occur: "In Sanskrit every word (except adverbs and particles) is inflected,
and the grammatical inflexion itself shows the. relation in which one word
stands to another. Thus grammatically speaking, there is no order as such
that need be much attended to .... But if there is no grammatical order,
there is a sort of logical sequence of ideas, which must follow one another in
a particular order. ... Words must be so arranged that the ideas will follow
one another in their natural order, and the words in their natural con-
nection .... " (Apte 1963, 263-4).
Though Apte was, among other things, the superintendent of a New
English School, and wrote his 'Guide' in English for Indian students, one
need not be surprised at finding that these reflect distinctions
current among San$krit grammarians at least from the time of Pataiijali's
i.e., for more than two millennia. The distinction which is
relevant in the present context is that between SaJfibandha 'the relation of
one word to another within a sentence (as shown, e.g., by grammatical
inflexion)' and abhisat[lbandha, anupiirvya or iinupurvi 'the order or ar-
rangement of words (as occurring in actual utterances)'. This distinction
could be specified within a general theory of language; in India it served to
delimit the scope of the science of grammar or vyiikarm;a itself. For, though
it is sometimes said that the Sanskrit grammarians were interested in grammar
but not in syntax, it would be misleading to interpret this as asserting that
they were interested in words (pada) and not in sentences (viikya) (cf. Renou
1960, 66; 1961, 129). It would however be correct to say that the Sanskrit
grammarians were interested in sa1flbandha and not in abhisaiJtbanda, and
in grammatical relations but not necessarily in word order.
The distinction between abhisa1flbandha and sa1flbandha will here be used
in order to study so-called 'free word order'. Despite appearences it might
1
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
be said, as we shall see, that the existence of this phenomenon is not clearly
established and that its significance has not been adequately clarified. 'Free
word order' itself (which would, if anywhere, obtain in a language like
Sanskrit) will here be studied not only for its own sake but also because
of its relevance to contemporary linguistic theory. This study will accordingly
be }?refaced with some remarks about recent developments in linguistics
showing what place is occupied by considerations of word order. At the same
time these remarks will provide the framework for the descriptions which
follow.
The Greek word for order, led not only to the formation of the
term syntax, but also to the framing of the term taxonomy, which at present
serves. to characterize the main inadequacy of structural linguistics. Lees,
who uses the term in this linguistic sense, described taxonomy as "the view
that sentences are constructed from left to right, word for word, by the
simple adjunction of successive constituents one to another" (Lees 1963
2
,
xix). Taxonomy, in this sense, corresponds to the Sanskrit abhisarrzbandha,
and structural linguistics can in this sense be characterized by its insistence
upon abhisarrzbandha. But structural linguistics may be judged not only by
this emphasis on taxonomy, but also by the "idea of basing a taxonomy
strictly and exclusively upon phonemic form" (ibid., xxi). In general, taxo-
nomic linguistics insists on confining its attention to the phonemic form,
arrangement and intonation of utterances by classifying their elements in
terms of simple labelled bracketing. The utterances thus analysed are col-
lected from a fixed corpus. All this is postulated in order to Gonfer on lin-
guistics the status of an empirical science. Aciually, such empiricism is
misguided (cf. Chomsky 1964, as well as several studies by P. M. Postal).
This misguided form of empiricism could easily combine with a behaviourist
theory of language use and learning. At the same time it adhered to the idea
of a discovery procedure by means of which a grammar is supposedly
obtainable, by mechanical means, from the physical shape of the utterances
obtained from a finite corpus (just as parts of botany are presumably ob-
tainable, by mechanical means, from a stretch of jungle).
Just as structural linguistics may be characterized by its insistence upon
taxonomy, the Sanskrit grammatical tradition may with reference to syntax
be characterized by its insistence upon sarrzbandha or grammatical relations.
This has not been generally recognized; even Lees referred to "the
or Taxonomic approach" in connection with the treatment of compounds
(Lees ibid.). This characterization may be due to the particular interpretations
Pli!fini has undergone at the hands of some [.9th-century Western linguists.
Whitney, for example, was not only a Sanskrit scholar to some extent
familiar with but has also been quoted (by Chomsky) as one who
2
INTRODUCTION
expressed most clearly the taxonomist point of view, namely, that "language
m the concrete sense ... (is) ... the sum of words and phrases by which any
man expresses his thoughts" (quoted Chomsky 1964, 22). The same Whitney
expressed with regard to Pai)ini's grammar his disapproval of "the highly
artful and difficult form of about four thousand algebraic formula-like rules
in the statement and arrangement of which brevity alone is had in view, at
the cost of distinctness and unambiguousness" (quoted from Whitney's
grammar by Buiskooll939, 2, who subsequently refutes this view).
The correct approach to the study of the Sanskrit grammarians was
outlined by Faddegon, whose description in 1936 of "the Occidentallinguists
of the latter half ofthe nineteenth century" applies equally well to present-day
taxonomists. It is clear, moreover, that Faddegon thought especially of
Whitney in this connexion (cf. a phrase from the motto for his 1936 book:
"Whitney has admirably attacked Pai)ini but for Pai)ini we forget Whitney").
Referring to the ktiraka theory, Faddegon said: "Evidently Pai)ini tries in
this analysis to separate the ideational aspect from the linguistic expression,
an attempt which the Occidentallinguists of the latter half of the nineteenth
century have condemned, misled as they were by the hope of being able to
understand language through the exclusive study of its phonal and morpho-
logical aspect, i.e. its articulative utterance and the association-system under-
lying declension and conjugation, as if the application and imitation of phys-
ics and a mechanistic psychology were the last word of moral science. And
so besides the injustice done to a pioneer of grammar who lived about
twenty-five centuries ago by associating with him results of niodern grammar,
it is even questionable whether Pai)ini has not something still to say to us"
(Faddegon 1936, 18).
We shall repeatedly be in a position to observe that Pai)ini continues
to have something relevant to say. But let us return to the framework of
modern linguistics that will be adopted here. The development from structur-
al linguistics to generative grammar or what could be called 'theoretical
linguistics' can be most readily described by stating (I) _that the generative
grammarians discovered (or re-discovered: see below) the fundamental
importance of rules for the statement of linguistic regula;ities; and (2) that
for the generative grammarians, the observations of structural linguistics
relate exclusively to the surface structures of sentences. This follows from
the discovery (or re-discovery, to some extent: see Chomsky 1966a) that
interpretations in terms of these surface structures are inadequate (e.g.,
in the simplest case, for describing constructional homonymy) and that
assumptions have to be made (like mutatis mutandis in almost any other
science) with regard to abstract structures underlying the surface structures.
Such underlying structures constitute a deep structure, which directly deter-
3
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
mines the semantic interpretation of sentences. The deep structure is de-
scribed with the help of a relatively simple kind of re-writing rules, which
are said to generate the base; when formulating such rules, we shall make use
of a single arrow (--> ). Rules of a different type, i.e., transformational rules,
ar" needed to convert deep structure into the surface structures which
the physical shape of actual utterances; when formulating such
rules, we shall make use of a double arrow ( => ), which will also be used in
formulating context-sensitive rules.
While natural languages differ greatly in their surface structure, the
difference in deep structure is, to say the least, less obvious. This has led in
turn to the further assumption that the deep structure, which determines the
semantic interpretation which must to some extent be invariant
translation, largely incorporates what was to be a class oflinguistic universals,
i.e., universal structures valid for all natural languages, or structures of
'universal grammar'. This has linked these recent developments with tradi-
tional theorles of universal grammar and with classical European rationalism
(Chomsky 1966a), and at the same time has made for a much closer connec-
tion between linguistics, psychology and philosophy- itself again constituting
a return to an earlier point of view (ibid., 76, note 4).
In this sketchy survey of contemporary linguistic theory one important
element has so flir been omitted: the incorporation into linguistic theory of
rules which are able to account for the fact that there are infinite possibilities
of expression in every language. The main technical contribution of generative
grammar in this respect is the introduction (from logic and mathematics)
of recursive rules, which safeguard the possibility of deriving infinitely many
expressions. After such rules had first been introduced into the transfor-
mational component of the syntactic pat! of generative grammar, they were
subsequently confined to the base which expresses the deep structure. The
simplest solution which is nowadays adopted is the incorporation of a
sing]el recursive rule in the base component (excepting a rule schema
required for conjunction). This rule is of the form:
where S is the initial category symbol which occurs on the left of the arrow
in the initial rule of each derivation, 'n' expresses concatenation and A is
a category symbol which is introduced by earlier rules of the derivation.
The recurrence of S guarantees the possibility of derivation of infinitely
many expressions. All this will be illustrated below.
It may be noted that the notion of rule corresponds to the notion of
1 Actually, a pair of recursive rules: (3a) and (3b) on page 6 below. For a recent
fication see Rosenbaum 1967.
4
INTRODUCTION
siitra as it occurred iu Pat;tini.
2
His commentator Pataiijali (traditionally
placed in the 2nd century B.C.; cf. below, page 48 note 19) was explicit
with regard to the fact that only a finite number of rnles are required to
account for infinitely many expressions.
3
The recursive character of such
rules is implicit in the comparison made by the 5th-century philosopher-
grammarian Bhartrhari between the recognition of new sentences and
counting.
4
In what follows we shall adopt a simplified version of the rules of the base
as given in Chomsky's Aspects.ofthe Theory of Syntax (1965). This simplified
version (in which sentence boundary marks, for example, will always be
omitted) is clearly insufficient for the description of all sentences, e.g.,
of English, but it already provides for the description of infinitely many
English sentences and it appears to possess all the characteristics of the system
that will be required in the following discussion. Though this system is now
accessible in a published form, its presentation in Chomsky 1965 is combined
with other investigations. A relatively detailed explanation will therefore be
given of the simplified version adopted here, so that precisely the framework
which provides the background for the following discussion will become
available. (The rules of the base of Chomsky 1965 have already undergone
some modification in more recent work; cf. e.g. Rosenbaum 1967. But this
is also inessential in the present context).
The re-writing rules of the base, which expresses the deep structure, are the
following (cf. Chomsky 1965, 106-7):
(1)
2
See, e.g., Staal19.63; for siitra in gerieral see Renou 1963.
3
Mahiibhii$ya, ed. Kielhorn 5.23-6.3 original is quoted in Staal 1967b, note 1):
"Now if grammatical expressions are taught, must this be done by the recitation of each
particular word for the understanding of grammatical expressions- must, e.g., the gram-
matical expressions 'cow', 'horse', 'man', 'elephant', 'kite, 'deer', 'Brahman' be recited?
No, says the author, this recitation of each particular word is not a means for the under-
standing of grammatical expressions. For there is the following tradition. Brhaspati
addreSsed Indra during a thousand divine years going over the grammatical expressions
by speaking each particular word, and still he did not attain the end. With Brhaspati as the
instructor, Indra as the student and a thousand divine years as the period of study the end
could not be attained, so what of the present day when he who lives a life entirely lives at
most a hundred years? ... Therefore the recitation of each particular word is not a means
for the understanding of grammatical expressions. - But then how _are grammatical
expressions acquired? Some work containing general rules and exceptions has to be com
posed . ... "
4
Vakyapadiya 1.87: yathiidyasalikhyiigrahavam upiiya(l pratipattaye smikhyiintariiQiim
bhede'pi tathii SabdiintaraSruti/:t 'just as grasping the first numbers is a means fOr the under
standing of other numbers, even when different, so is hearing other words'. The corn
mentator explains the last words as follows: tathii devadattiidiSabdiintara
Srutiparicchedopiiyii viikyariipaprati'patti/:t 'so a sentence is understood by means of analysis
from hearing other words, as Devadatta'.
5
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
VP--> V"NP
NP--> (Det)"N':'(S).
(2)
(3)
The parentheses in the third rule denote option, i.e., (3) stands for the
following four rules:
)
'
NP--> Det"N"S
NP--.N"S
NP--> Det"N
NP--.N.
(3a)
(3b)
(3c)
(3d)
Thus (3a) and (3b) are the only rules which contribute recursiveness. The
symbols may be read as follows:
S: 'Sentence'
NP: 'Noun Phrase'
V: 'Verb'
VP: 'Verb Phrase'
Det: 'Determiner'
N: 'Noun'.
These rules permit the construction of infinitely many trees, e.g.:
s
~
Det N
s
NP
~
Det N S
N ~ P
A A
(i)
)\
Det N
)\
V)\
Det N
Det N V )\ (ij)
etcetera.
Det N
6
INTRODUCTION
We shall now, greatly simplifying Chomsky's conventions (1965, especially
79-111; cf. a!s<;> Staal 1967a), adopt a dictionary consisting in the following
English vocabulary:
(book, N)
(enter, V)
(own, V)
(student, N)
(the, Det)
(university, N).
Then, in accordance with Chomsky's conventions and adopting rules as
illustrated by the tree (i), we can derive such strings as: 'the book enter the
university', 'the student enter the university', 'the s t ~ e n t own the book',
'the student own the university', 'the student enter the student', 'the university
own the book', etc. Transformational rules will be required to introduce
concord or government into these strings (the actual shape of some of these
rules will be considered later) and, moreover, certain transformational
markers will have to be introduced into the base. Thereby the following
sentences can be derived: 'the student enters the university', 'the students
enter the university', etc. Certain sentences may be eliminated provided
restrictions of a semantic nature are built into the dictionary. Thus we
may eliminate, e.g.; 'the university enters the student', if so desired.
The tree (ij), on the other hand, symbolizes the derivation of strings that
are more interesting, e.g.: 'the student the student own the book enter the
university' or 'the university the university own the student own the book'.
Other transformational rules are now required to derive sentences from these
strings, e.g. (from the first) 'the student who owns the book enters the uni-
versity' (but also: 'book owning students enter the university'), and (from the
second) 'the university which owns the student. owns the book' (and
perhaps: 'the university which owns the student owns his book').
With the help of such derivations as symbolized by the trees (i) and (ij)
a great many sentences, but not infinitely many, can be derived. By con-
structing trees on the basis of(l)-(3) in which (3a) and (3b) are introduced
c;.gain and again, we can derive as maay sentences as we wish.
In principle this is a remarkably simple and, as can be seen from the
regults, extremely powerful method of description. Though the form of
many specific transformational rules, and also of transformational rules in
general, remains for the time being a matter of some uncertainty or dispute,
the overall structure of the system seems clear. Its appropriateness, in princi-
ple, for the generation of English sentences is partly based upon a specific
feature which characterizes Chomskian generative grammar: i.e., the fact
that order is not only a feature of the terminal strings and finally derived
sentences (as certainly it shol)ld be), but also figures in the deep structures
underlying them. The reason for this is that the re-writing rules of the base,
e.g., (1)-(3), introduce ordered strings by making use of concatenation.
7
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
Accordingly, the trees have a fixed structure from left to right as well. At
first sight, this incorporation of order into the deep structure may seem an
unfortunate and even paradoxical inheritance from structural linguistics.
For the taxonomy of structural linguistics was rejected exactly because of its
a<\herence to surface structure, in which order plays a paramount role.
So\why preserve order in the deep structure? Indeed, if order were rejected,
one could adopt instead of the rule (I), which introduces an ordered string,
a rule (I'), which introduces an unordered set, i.e.:
S-+{NP, VP}, (!')
where {NP, VP} is defined as the set which contains the two elements
NP and VP; hence {NP, VP} and {VP, NP} are different expressions
denoting the same thing. With the help of(!') one could, however, derive
the following strings: 'own the university the student', 'enter the university
the book', etc. In English, at least, this would lead to an entirely redundant
proliferation ofungrammatical and unwanted strings.
It is obvious that order is often further determined by transformational
rules. For example, we derive: 'the students who own books enter the uni-
versity' as against 'the book owning students enter the university', and not:
'the who own books students enter the university' as against 'the students
book owning enter the university'. It may therefore be asked whether there
is any good reason for not selecting, to the right of the arrow in (3), another
o ~ e r e string, i.e.:
NP-+ (Det)'"'(S)"N. (3')
So far there is no criterion which enables us to choose between (3) and (3'),
though, of course, the transformational rules required in addition will have
to be formulated differently in the two cases.
On the whole, and excepting the case just mentioned, it seems justified to
introduce the given order into the deep structure, at least for the purpose of
deriving sentences in English and in oiher languages with a similar surface
structure. This means, among other things, that the deep structure of
Chomsky's generative grammars is not similar to what the Sanskrit gram-
marians called sal]7bandha 'the relation of one word to another' (as shown,
e.g., by grammatical inflexion), since these grammatical relations have noth-
ing todo with surface order. There are however grammatical relations which
can be derived on account of such trees as"(i). Chomsky defined grammatical
relations as ordered pairs in the following way (1965, 71):
. 8
Subject-of: [NP, S]
Direct-Object-of: [NP, VP]
Main-Verb-of: [V, VP] etc .
(4)
(5)
(6)
INTRODUCTION
It may be noticed that these relations can be defined in terms of the deep
structure without making use of its ordering: (4)-(6) have just the same
significance in a system by which differently ordered strings are introduced,
e.g.:
S _,. VP"' NP
as they have in a system by which unordered sets such as (I') are introduced.
Moreover, the same grammatical relations could also be defined, if other
constituents were to intervene in the ordered strings which are derived, e.g.,
if we would for other reasons want to adopt expressions such as:
S .... NP"'Adverbial"'VP.
The system of grammatical, relations defined by (4)-(6) is much closer to
the grammatical functions and relations which occur in traditional parsing
in Western grammars, and is at the same time much closer to what the
Sanskrit grammarians called sa111bandha.
In order to find out to what extent the deep structure reveals linguistic
universals one might investigate what happens when the following (again
greatly simplified) Sanskrit vocabulary is adopted in addition to the sinapli-
fied base consisting of (I )-(3):
(apasyat, V)
(govinda, N)
(rama, N).
For the sake of simplicity the form apasyat 'he saw' has been chosen as the
representative form of the Sanskrit Verb for 'see'. With tree (i) we can Iiow
derive: riima apaSyat gavinda and govinda apaSyat rtima. Transformational
rules introducing concord or government and sandhi will be required to
derive from these two strings the sentences rtimo 'paSyad govind01n 'Rama saw
Govinda' and govinda 'paSyad rtimam 'Govinda saw Rli1na', respectively.
The derivation so far appears to be largely similar to the derivation of
the English sentences 'the book enters the university' and 'the university
enters the book'. But this similarity is somewhat misleading since it arises
from our failing to correctly specify the required transformational rules.
The English example could have been dealt with by adopting any transfor-
mational rule which effectuates the affixing of '-s' to 'enter-' on account of
the fact that 'the book' is a Singular NP. This could, for example, be done by
means of a context-sensitive rule of the form (9), following a base consisting
of (I), (7), and (8)
5
:
S--;NP':'VP
(1)
5
Cf. Chomsky 1957, 28-9; Lees 1963
2
, 45; Chomsky 1965, 175-6.
9
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
NP-> (Det)"(S)"W'{!n
(7)
VP->NP"V"{!n
(8)
)
N"{!n"VP
(9)
The braces in the rules denote alternatives, e.g., (9) stands for the following
two rules:
N"Sg"VP N"Sg"NP"V"Sg
N"Pl"VP
(9a)
(9b)
where Sg and PI denote Singular and Plural, respectively. In addition we
shall require some transformational rules which effectuate:
book"Sg
enter,-...Sg::::::;::. enters.
As for the Sanskrit examples, a similar though slightly more complicated
solution would seem fit. First of all, any more general description must
contain a rule which is analogous to (9). This has been avoided artificially
in the present example by adopting apayat as the representative form of the
Verb. But in addition, to take the first example, rtima requires a Nominative
Case ending and govinda requires an Accusative Case ending. This could be
achieved, e.g., by context-sensitive rules (12) and (13) following a base
consisting of (1), (8), (10), and (11):
NP-> }"Case (I 0)
Nom
Ace
Instr
Case-+
Dat
Ablat (11) 6
Gen
Loc
Voc
6
Note that (11) might have been incorporated into (10) in the same manner in which
is incorporated into (7)-(9).
10
INTRODUCTION
}"Case"VP=> N"{:: }"Nom "VP
N"{Sg}"Case"V => N"{Sg}"Acc"V.
PI PI
In addition we require transformational rules which will effectuate:
rtimanSg-'"'Nom => rtimal;z
govindar.sgnAcc=> govindam.
(12)
(13)
This enables us to derive the string ramab apasyat govindam and from this
in turn, with the help of transformational sandhi rules, we derive the first
sentence, i.e., riimo 'payad govindam.
For the present purpose we need not enquire whether the above solution
is the best possible (transformational rules might for example have to
be used instead of context-sensitive rules); this solution merely serves to
demonstrate that a solution can be found, which then turns out to be some-
what more complicated for the Sanskrit than it was for the English example.
It is worth noticing, however, that the context-sensitive rules which have
been utilized, though formulated in terms of ordered strings, are in fact based
upon grammatical relations which do not depend on any ordering. As
regards English, (9) expresses that the Number of the Verb of the VP is
governed by the Number of the Subject of S; but nothing is presupposed
with regard to the order of these elements. As regards Sanskrit, (12) expresses
that the Subject ofS has the Nominative Case ending, and (13) expresses that
the Object of VP (or, equivalently, of S) has the Accusative Case
again, the order of elements is irrelevant This implies that we introduce
redundant information by introducing rules, which presup-
pose order, for expressing facts of concord or government which have
nothing to do with order. The question arises whether this should, and how
it could be avoided.
We shall return to this problem much However, it may be observed
that the redundant information carried along by the context-sensitive rules
is exactly the information required for establishing the proper word order in
English. But it is precisely here that the apparent parallelism between English
and Sanskrit breaks down; for in Sanskrit the situation is different. It seems
difficult to question the fact that ramo 'pasyad govindam is not the only
Sanskrit sentence which means 'Rama saw Govinda'. In this connection
Apte may once more be quoted. On the first page of his 'Gnide' he says:
"Take ... the English sentence 'Rama saw Govind'. If the order of words,
'Rama' and 'Govind' be changed, there will be a very great difference in
the meaning; it will, in fact, be a different sentence altogether. Take, however,
11
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
the Sanskrit sentence for the same: ramo govindam apasyat. Here, even if
the order of the words be changed, no difference occurs in the meaning: the
sentences rtimo govindam apaSyat, govindalfl riimo 'paSyat, apaSyad riimo
govindam&c., all mean the same thing" (Apte 1963, 1).
Since transformations do not change meaning (Katz &Postal 1964; cf.
1965b) and are, at least in a generative grammar of English, in general
used to re-arrange the order of words, it appears to stand to reason that trans-
formational rules will be postulated to account for the required permutations.
It might pay to be careful and first adopt a base which directly generates the
sentence which Apte quoted first, i.e. ramo govindam apasyat. This can be
derived from:
S -+NP"'VP
VP-+NP"'V.
(14)
(15)
Subsequently three transformational rules will be added, which will be
defined appropriately but in a simplified way (omitting the dominating
constituents of the Phrase-marker) for the tree generated by (14)-(15), i.e.:
NP"'VP= VP"'NP
NP"'V =V"'NP
V"'NP =NP"'V.
(16)
(17)
(18)
These rules have to be optional, since, for example, (17) and (18) would
otherwise cancel e&ch other and neither (16) nor (17) should be applied in
the derivation of ramo govindam apasyat. Then govindaJ?1 ramo 'pasyat can
be derived by applying (14), (16), (15) and (18); and apasyad
ramo govindam can be derived by applying consecutively (14), (15), (17), and
again (17). The three remaining possibilities (presumably hinted at by Apte's
'&c.') can be similarly derived.
It is obvious that these manipulations are far from elegant. Chomsky
therefore has, with reference to stylistic inversion (which is superficially
similar to the syntactic inversion described here), emphasized "that gram-
matical transformations do not seem to be an appropriate device for ex-
pressing the full range of possibilities" (1965, 126). But even if other
rules, similar to (16)-(18), have to be adopted, this will not only greatly
detract from the simplicity of the description, but also give preferential
treatment to one of the six permutations (in the present description, the
order generated by (14)-(15)) for which a specific syntactic justification
will be needed.
Actually, the general situation is more complicated. Chomsky has sug-
gested that "stylistic inversion of 'major constituents' (in some sense to be
defined) is tolerated up to ambiguity" (1965, 127)- the excluded ambiguities
12
INTRODUCTION
being of the type of Hirt's example from German "Die Mutter sieht die
Tochter" (see, however, below, page 68). But unless V is included among
major constituents, rules such as (17) and (18) will not be admitted and this
would prevent us from accounting for the six permutations considered above.
Moreover, the required permutations need not generally be confined to
elements of substrings dominated by major constituents, as we shall see
below (pages 34, 54). Furthermore, the objection that preferential treatment is
given to one particular permut:ition remains valid also when some general-
ization about permutability is postulated. Generalizations about permuta-
bility, then, whether to account for stylistic or for syntactic inversion, seem
to be quite complicated ad hoc hypotheses, necessitated by the earlier
adherence to strings. If 'free word order' should constitute a basic linguistic
fact, this state of affairs would therefore suggest a description in terms other
than strings - at least to start with. Word order would, of course, have to be
introduced, but at a later stage (e.g., after affixing Case terminations, but
before applying external sandhi rules).
The connection between inflexion and free word order has long been
known in the West, especially on account of the comparison of the vernacular
languages with Latin, which in this respect is at least to some extent similar
to Sanskrit. Chomsky has quoted statements about this connection from Du
Marsais and from Adam Smith (1966, 45, 101). But even the view that word
order does not belong to what is nowadays called the deep structure, has
been maintained before. Du Marsais distinguished between 'construction' and
'syntaxe' (Chomsky 1966, 47). Construction is applied to "]'arrangement
des mots d'lns le discours", >yntaxe is applied to "les rapports que les mots
ont entre eux". Chomsky noted that "rhe syntax of an expression is thus
essentially what we have called its deep structure; its construction is what we
have called its surface structure" (ibid.). In a footnote (88, on p. 103) he
refers, however, to the discussion in Aspects, w h e ~ e as we have seen, word
order does essentially belong to the deep structure. We shall return to this
question, but it may be noted here that the distinction Du Marsais made
between syntaxe and construction corresponds to the distlliction of the
Sanskrit grammarians between sal]'lbandha and abhisal]'lbandha.
Recently Curry (1961), Hii:7 and the present writer (1965c, 178 note;
1966a, 192; 1966b, 251) have accepted free word order as a linguistic fact
and have criticized generative grammar on this account. Curry has suggested
that grammatical structure be studied in terms ofjunctors. He proposes to call
the level of grammar which studies grammatical structure: tectogrammatics,
and the level of grammar which studies how grammatical structure is repre-
7
In: Structure of Language and its Mathematical Aspects, Providence, R.I., 1961, 265.
13
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
sented in terms of expressions: phenogrammatics. These terms can now
be shown to correspond to deep structure and surface structure, respectively,
provided order is excluded from the deep structure. For, as Curry observes
(1961, 65), "if phrase structure grammar means the building up of phrases by
concatenation of adjacent h r a s ~ s then it has a phenogrammatical aspect".
WhUe it seems likely that many aspects of the deep structure can indeed be
described in terms of functors, it is not clear what kind of rules and mecha-
nisms would be needed in phenogrammatics in order io derive surface
expressions from grammatical structure, unless familiar transformational
rules are adapted for that purpose. The use of functors in tectogrammatics
for the description of deep structure, however, deserves to be further ex-
plored.
The theories which have been worked out in much greater detail by
Saumjan and Soboleva s are at first sight similar. Saumjan distinguishes
between two levels of linguistic description, the genotype and the phenotype.
level. Word order is again excluded from the genotype level. But while it is
again not clear how in a number of cases the transition from genotype to
phenotype is made, Barbara Hall has in addition shown (Hall 1964) that
Saumjan's description of the genotype itself is unsatisfactory in several ways
(e.g., the members of the well-known pair 'John is easy to please' and 'John is
eager to please' - Chomsky 1964, 34-5, 60-5 - come out with the same
grammatical structure). It stands to reason that Chomsky has therefore
strongly objected to such alternatives (1965, 124-5) and has furthermore
emphasized that the introduction of sets (as in (I')) instead of strings does not
solve any difficulty and in fact creates new problems, since such sets will in
turn have to be replaced by strings before actual sentences can be finally
derived. The imperfections of existing alternatives to Chomsky's theory,
however, do not invalidate the criticisms motivating them. It remains, for
example, counter-intuitive to assume that rules of concord must be formulated
in terms of strings, though concord itself has nothing to do with order.
Moreover, even if order is introduced at a later stage, the rules which would
precede this introduction might have a much wider scope and higher degree
of generality. Such rules could in fact be more readily regarded as rules of
a universal grammar.
It may be noticed that rules introducing sets, e.g.:
S ->{NP, VP}
VP->{NP, V}
NP-> {Det, S, N}
s See Saumjan 1965 and the literature mentioned ibid. 191-2, note 4.
14
INTRODUCTION
not merely lead to the same system of grammatical relations as rules
introducing ordered strings do, but can also very well combine with such
trees as (i) and (ij), provided these trees are interpreted in a different manner,
i.e., without postulating that the branches branch off in a specific order.
We may accordingly consider wild trees' as configurations of points such
that there is one point at the top and each other point is connected with one
and only one point above it. Trimmed trees, on the other hand, are wild
trees with their branches branching off in a specific order, i.e. trees of the
type now familiar in the literature on generative grammar. In wild trees the
points dominated by a node constitute a set; in trimmed trees the points
dominated by a node constitute an ordered sequence. As noted before,
grammatical relations (and also salflbandha) can be defined for wild trees as
well as for trimmed trees. On the other hand, since wild trees are especially
appropriate for symbolizing rules which introduce sets, context-sensitive
rules cannot be defined for wild trees. But again there is no difficulty in
defining the corresponding rules for sets, which could be called company-
sensitive rules.
There is no point in introducing wild trees, company-sensitive rules and
other concepts which are unfamiliar to linguists unless it is fairly probable
that there is such a thing as free word order to account for. Moreover, this
phenomenon itself, once found and analysed, may suggest specific methods
for its adequate description. The following investigation into free word
order will be mainly confined to Sanskrit. This has the additional advantage
that Sanskrit has not only been the object of Western linguistic and phi-
lological scholarship for at least one century, but has also been the object
of Sanskrit scholarship for at least two millennia. In the second section,
Sanskrit word order will" accordingly be considered as it has been studied by
the Sanskrit grammarians and theorists of language. The third section will
consider the same topic as studied by Western Sanskritists. The fourth and
last section will give a summary and formulate conclusions from the pre-
ceding two sections, which at first sight seem strangely incompatible; it will
seek to study methods of linguistic description that can deal adequately
with the material studied; and finally it will seek to contribute to the solution
of some of the general problems mentioned in the present Introduction.
9 Note that what are here called 'wild trees' are a variety studied by mathematicians: see
Berge 1958, chap. 16, and Ore 1962, chap. 4 (I owe these references to Barren Brainerd).
15
CHAPTER li
THEORISTS ON WORD ORDER IN SANSKRIT
)
'
The term 'Sanskrit grammarians' is generally used to refer to the adherents
of the science of vytikarava (literally, 'analysis'). They may be grouped into
several schools, the most important of which is the school of Pal).ini. The
present problem however requires us to make some excursions beyond the
confines of vyakarava proper into the linguistic speculations first of the
Vedic ritualists and second of the ritualist philosophers of the Mlmamsa.
Members of any of these three groups will be referred to as 'Indian theorists'
or 'Sanskrit theorists'. Some attention must now be devoted to each of
these three groups, to the sources that will be quoted later, and to the back-
ground of these sources. The three groups will be referred to as (a) Vedic
ritualists, (b) grammarians, and (c) Mlmiimsii philosophers.
(a) The Vedic ritualists evolved the siitra style which constitutes a unique,
and at the same time one of the most characteristic manifestations of Sanskrit
literature (for an excellent survey of this style see Renou 1963). The term
satra or 'thread' denotes a rule or aphorism, or a teXt which consists of rules
or aphorisms; in the latter sense the term is generally written with a capital,
as Siitra. These rules are often formulated in a very concise manner. This
is, for example, the case in Pal)ini's grammar, where the siltras come to resem-
ble algebraic formulas (as noted already by Whitney, quoted above, page 3;
for further details see Staal 1965a). The odd combinations that may result
from such concision are referred to in the following pun on the double
meaning of siltra (quoted by Staall963a, 32 = 1965d, liS):
'No wonder that the girl strings together glass, gems, and gold on
one thread.
Even Piil).ini, who ought to have known better, combined dog, youth
and king of the Gods in one rule.'
The earliest siitras were the ritual siitras. The authors of the still earlier
Brahmal).a literature had engaged in interpretations of the Vedic ritual and
in related speculations which developed further in a metaphysical direction,
e.g., in the The authors of the ritual siitras, on the other hand,
confine themselves to minute and painstaking descriptions of the details of
16
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
the ritual. One of the most important features of Vedic ritual is the recitation
or chanting of Vedic texts; this had become almost as complicated as the
ritual itself (see e.g. Staall961). Special sutra texts were therefore composed
to deal with the recitation of the Veda. The resulting literature is called the
Pratisakhya literature, and consists of a number of texts, in principle one for
each of the Vedicschools (as the name derived from sakhd 'school' indicates).
The PratiSiikhya literature may be placed in the period 500-150 B.C. The
probably earliest treatise is that of the J3.gveda, and is therefore called the
]J.kprdtisdkhya. It is written in verse but possesses sutra characteristics and
is referred to as such (see Renou 1963, 201, note 9). Apart from the
]J.kprdtiSdkhya, important PratiSiikbya texts are attached to the Black and
White Yajurveda. Similar texts, partly overlapping, belong to the Athar-
vaveda. The Samaveda, which consists in part of verses of the J3.gveda set
to music, possesses a number of related texts which deal e.g. with the corre-
lations between verse and melody and the connected correlations between
accents and tones. - The existence of other Pratisakhya texts may be
assumed.
The Pratisakhya literature deals primarily with the rules of euphonic
combination (sandhi). Their phonological content is described e.g. in Varma
1929 and Alien 1953. Special attention is paid in the Pratisakhya literature to
the analysis of syllables, accents, compounds, and the boundaries of words
and sentences. From these texts Indian linguistics developed. When it is
said !hat the study of n ~ u g e in India "was undertaken in the first place
primarily to meet the needs of Vedic ritnal and the text material required by
it" (Brough 1953, 161, and cf. Emeneau !955, 151) it is mainly the Pratisakhya
literature that is had in mind.
It is important to observe that the object of study of the Pratisakhya
literature was the corpus of Vedic texts, which was transmitted orally. This
implies that the authors of these texts were not interested in the Vedic
language as such, but in the utterances handed down in the Vedic corpus and
in particular in the Vedic school to which they belonged.
We shall quote from the probably earliest and most important ]J.kprdtisa-
khya, which was commented upon by Uvata(ofuncertaindate). We shall also
quote from the Nirukta, a work dealing primarily with Vedic etymology,
which may be assigned to an early date within the same period as that of
the Pratisakhya texts. The Nirukta was commented upon by Durga (13th
century?
1
). A later work which provides a list of Vedic deities but also
contains a grammatical section,. the Brhaddevatd, partly depends on the
Nirukta.
1 According to Sarup 1920, 50.
17
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
(b) The grammatical tradition antedates Piil).ini, but Piil).ini's grammar
(which refers to several of its lost predecessors) is the oldest extant, and
continues to be the most important and complete Sanskrit grammar.
It consists of eight chapters or adhyaya and is therefore generally called the
A$1iidhyayi. The chapters consist in all of about 4000 short rules called
Each chapter consists, moreover, of four quarters or pdda, and the
rules are numbered consecutively from the beginning of each pada. It is
therefore customary to refer to each rule with the help of three numerals, the
first denoting the chapter, the second the quarter of that chapter, and the
third the siltra of that quarter (e.g., I .4.80 for the 80th siitra of the fourth
pada of the first adlzyaya).
The object of Piil).ini's grammar was the complete description of Sanskrit
as the spoken. language of his time and, possibly to a more limited
extent, of the Vedic language (chandas) (Renou 1967,). The date of
Pa1.1ini is quite uncertain, though he is often placed in the 4th or 5th century
B.C.
2
It has been a matter of some dispute whether the grammar of Pa1.1ini
was earlier or later than the ]J.kpratisakhya (cf. below, page 21).
Whereas the question in how far the Vedic language (chandas) is dealt
with in grammar has also been much discussed, no special attention
seems so far to have been given to another fundamental difference between
the Pratisakhya literature and the grammatical tradition. Whereas the
authors of the former are, as we have seen, mainly interested in the corpus
of the Vedic texts, the grammarians were interested in the language itself,
whether Vedic or contemporary ('spoken'). The phonetic. treatises belonging
to the Atharvaveda may in this respect form a kind of transition between the
other Pratisakhya texts and the works of the grammarians (cf. Renou 1947,
82-3, 224-5). The general difference in outlook and method between Vedic
ritualists and grammarians implies that the former were mainly interested
in utterances, the latter in sentences. We have already had occasion to observe
that the grammarians stressed the infinity of the expressions of language
(above, page 5, and note 3). We shall furthermore see that they did
not hesitate to construct even Vedic sentences which were not known
from the corpus of utterances. The Vedic ritualists were not interested in
such constructions. (Incidentally, it may be noted that their lack of interest
was better motivated than that of those modern linguists who defend a
similar point of view on account of a certain misguided form of empir-
icism.)
Pa1.1ini's main commentator was Patanjali in his a voluminous
2
The literature on the subject is referred to by Renou 1967, and in other works by the
same author, quoted there.
18
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
work which is the main source for our interpretation of Piinini and for our
knowledge of early Indian linguistics and theories of language. The
may be assigned to the second century B.C.3 Pataiijali .also
comments upon Pa.J;rini's successor K3tyayana, whose 'rules of interpretation'
(vtirttika) are only known from the itself.
4
The Piininian
tradition continued in India - though not uninterrruptedly - with an im-
pressive lineage of commentators, often original scholars themselves (e.g.,
Nagojl- or Nagesa- Bhatta in the 18th century) and has remained alive to
the present day.
This is not the place to give a description even of the main features of
Piii;ini's grammatical system. This system will be frequently drawn upon
and illustrated in the following pages. It may be sufficient to state that
Piinini utilized a variety of technical terms and special elements of an arti-
ficially introduced meta-language, carefully distinguished from elements of
the object-language; that the economy criterion (cf. the often quoted maxim
"grammarians rejoice over the saving of the length of half a short vowel as
over the birth of a son") led him to adopt numerous special devices, in-
cluding anuvrtti 'recurrence' (of elements of earlier rules in later rules,
where they need not be mentioned explicitly) and, accordingly, a specific
ordering of the rules; and that special meta-rules occur among
the sfitras, the task of which is to lay down how the rules are to be applied
and how apparent inconsistencies are eliminated. It may be re-emphasized
that Piinini's grammar is not confined to the phonology and morphology
of Sanskrit (as has been suggested), but includes symax, and that with
reference to the littter it does not adopt a taxonomic outlook (as has been
suggested). We shall return to this below (pages 36 and following).
We shall quote from Piil)ini and Patafijali, and also from the 17th-century
Siddhtintakaumudi by Bha((oji ..
(c} The Mimiil]1sii constitutes a later development. In the Vedic literature,
the Briihmal).as and the may be said to deal to some extent with
ritual interpretation cum speculation, and general metaphysical speculation,
respectively. Two of the later systems of philosophy continue these two
trends in a more systematic and rationalistic way: the pilrva-mimiiJ?lSd
'prior investigation', karma-mimiiJ?lsii 'investigation into (ritual) activity' or,
briefly, Mimarpsa on the one hand, and the uttara-mlmiil!1Sii 'posterior
investigation' or Vedanta on the other hand. Whereas the Vediinta and its
extensive developments have attracted relatively much attention in the West
3
See Renou 1967, and below, note 19.
4
The relationship between Katyayana and Patafljali has been studied in exemplary fashion
by Kielhom (1876).
19
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
and continue to have a certain appeal, the seemingly arid prior Mlmatp.sa
contains in many respects more interesti.."l.g and original ideas - ideas,
moreover, for which it would not be easy to find counterparts outside the
Indian tradition. It has been realized that the Vedanta itself cannot be ade-
quately understood without a knowledge of the Mlmatp.sa ("Sailkara est tout
pen6tre de Mlmiimsa": Renou 1951, Ill). - For good introductions to the
Mrniatp.sa system see e.g. Jha 1942, Hiriyanna 1932, chapter XII, and cf.
Edgerton 1929 and Kuuhan Raja 1952.
The Mfmatp.sa has most probably developed from the ritual Sutras and
especially from the rules of interpretation attached to
some of these Siitras. But the Mlmamsa investigation is more general and
theoretical than the. investigations of the ritual Sutras. It was originally
realistic in outlook and technical in execution, and only later developed into
a full-fledged philosophical system. It combines trends which seem para-
doxical at first sight, e.g. an almost excessive concern with orthopraxy
'right activity' (as distinct from orthodoxy: 'right opinion') with a frankly
atheistic outlook. This is apparent e.g. from the doctrine that the deities
invoked in the course of the sacrifices are essentially grammatical datives. -
We shall see that the M1mar
1
1sa philosophers often return to the themes
which were discussed by the earlier Vedic ritualists.
Given the importance of the Vedic texts for the Vedic ritual, it stands to
reason that the Mlmatp.sa philosophers devoted much speculation to the
linguistic phenomena of the Veda. Their relation to the grammatical tra-
dition ' deserves special comment. Though these philosophers, like their
predecessors, dealt with Vedic utterances rather than with the sentences of
classical or spoken Sanskrit, their researches were motivated by a very
original theory about the nature of these utterances (cf. Edgerton 1928,
Staall962). Because of this and especially in the field of syntax, Mlmatp.sa was
in a position to supplement the investigations of the grammarians. Renou
has characterized the difference between vyakarava and Mlmatp.sa as that
between padamfmiif]1Sii 'investigation of words' and vdkyamfmdf{1sii 'in-
vestigation of sentences' (1960, 66; 1961, 129). It is true that the grammarians
were not interested in the surface arrangements of words in sentences;
but they were interested in external sandhi on the one hand, and in the
abstract underlying structures of sentences called sarrzbandha on the other.
While external sandhi is a secondary phenomenon in many other languages,
'syntax' has in the study of modern languages come to denote the study of
the superficial arrangements of words, as indeed the etymology of the term
'syntax' itself suggests. And so it is no more than correct to say that the
Sanskrit grammarians were not interested in sentences in as far as sentences
are considered merely as arrangements of words. But it would not be
20
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
correct to say that they were not interested in sentence construction. :s
The philosophers of the Mimiif[lsii look upon the Mimiif[lsii-siitra at-
tributed to Jaimini as their oldest source; beyond assigning this work to the
period 300 B. C.- 300 A. D., determination of its date seems hazardous. The
first commentator was Sabara, whose Mimiif[lsiisiitrabhii$ya is sometimes
assigned to the 5th century A.D. It is in turn commented upon by many
others, including the 8th-century scholar Kumarila Bhat!a in his Tantra-
viirttika as well as in other works.
As may be seen from this short survey the chronological and historical
relations between these three groups are sometimes hard to ascertain.
Though the third group is in some respects a continuation of the first, or
rather of the purely ritual siitras which accompany works of the first, the
first group deals on the whole with older material than the second. Ac-
cordingly many scholars place Pal)ini after the IJkpriitisiikhya and the
Nirukta, though Thieme has argued that Pal)ini must be earlier (Thieme
1935). It must be emphasized, finally, that many other important sources
(e.g,, non-Pal)inian works) have not been mentioned because they will not
be quoted below. It may also be worth noting that while we are often dealing
with the work of commentators, this does not imply that no important
original thought may be found in it ..
Word order will now be studied as dealt with in these works, and under
three headings: (A) J!.kpriitiSiikhya, Nirukta, Brhaddevatii; (B) Panini and
Patafijali; (C) Jaimini, Sabara and Kumarila Bhatta.
A. e.KPRATISAKHYA, NIRUKTA, Be.HADDEVATA
Most of the Vedic texts have been handed down orally in iwo traditional
forms: the SGf!1hitdptifha 'continuous recitation' and the padapiitha 'word for
word recitation'. The composition of the padapii{ha at a very early date con-
stitutes the first of a series of extraordinary precautions taken to preserve the
sacred text from loss or corruption. The saf[lhitiipii{ha and the padapii{ha
differ from each other principally with respect to external sandhi, nominal
composition and accentuation. An example from the B.gveda (10.127.2) is
the following:
Saf[lhitiipii{ha: I 6rvaprii timartyii nivtito devy udvtitab I
Padapii{ha: Id I uru I apriib I timartyii I nivtitab I devi I ut'vtitab I
'the immortal goddess has pervaded the wide space, the
depths, and the heights'.
5
Cf. Apte 1963
25
, 1: "'Syntax' in English deals with the mode of arranging words in
sentences, and lays down rules for the proper and correct arrangement of words. In Sans-
krit and other languages that are rich in inflexions, Syntax has not this definite scope."
21
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
The initial task of the Priitisiikhya literature is to teach how the sal!lhi-
tiipa{ha may be established from the padapa{ha. The given order of words is
explicitly recognized as the point of departure, e.g., in ]J.kpratisakhya 2.1
which states: anupiirvye!Ja sandhin '(one should make) the sandhi combination
according to the word order'.
oral recitation it is easy to forget one or two words from the padapatha.
The ]J.kpratisakhya therefore described another mnemonic device, called
kramapa{ha 'serial recitation'. If the padapa{ha be written: j a j b j c / d / ... ,
the kramapti{ha becomes:
/a bjbc/cd/ ...
For example, the above line yields:
Kramapatha: I 6ru I urvaprab I apra amartya I amartyti nivtltab I nivtito devi I
devy udvatab I
The kramapa{ha is the latest modification recognised at the time of the
]J.kpratisakhya.' But other modifications (vikrti) were developed later (e.g.,
ja{a: /ab/ b a/ ab// b c / c b / b c /I c dId c / c d //), and many continue
to be recited in different parts of India up to the present day (Staarl961).
Special devices were introduced to mark a compound word or the end
of a sentence. In its eleventh chapter the ]J.kpratisakhya gives rules for the
construction of the kramapa{ha from the sa1!7hittipa{ha. Jn a few cases
special problems arise on account of the order of words. In B.gveda 5.2.7,
where a reference occl'!s to the fe.mous myth of Sunal)sepa (literally 'dog
penis'), who was bought as a sacrificial victim at the price of a thousand cows,
the two traditional versions are at variance:
I nfditGI/1 sahtisrtid yilpiid amuFico hi
I evdsmcid agne vi mumugdhi pdsan ... j
Padapa{ha: I sunabsepam I cit I ni'ditam I sahtisrtit I yilpat I amuFicab I
I hi I sab I evtl I asmat I agne I vi / mumugdhi !
pdsan f ...
'Even Sunal)Sepa, who wao tied for a thousand, you have
released from the sacrificial post, for he was arrested; so
release us too from our fetters, o Agni .... '
If the kramapa{ha were constructed from the padapa{ha according to the
rules, the result would be:
I suna(1Sepal!1cit f cid nfditam j nidital!1 sahtisnit j ...
6
The ]J.kpriitiSiikhya asserts (11.66) that the krama "does not accomplish anything else"
(anyasiidhako na); this may indicate that other vikrtis were already current, though not
officially recognized at the time. These vikrtis and other specimens of Vedic recitation are
available on two LP records 'The Four Vedas' (Ethnic Folkways Library).
22
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
But from this result the order of words of the Saf[lhitiipiitha could never be
derived. f!.kpriitiSiikhya 11.13 proposes the following solution: .aniinupiirvye
padasandhyadarsaniit / padavyavetaf[l ea padaf[l vyaviiyi ea 'if the word order
is not in accordance (with the padapii{ha), the word divided by another and
the dividing word are passed over, since otherwise the sandhi of the words
(of the saf[lhitiipiitha) is not shown' (the word atiyante 'are passed over' has
to be supplied from 11.3, which has atiyate 'is passed over': a technique
well known from Pa11ini '). That is to say, both suna/;sepam and eit are
omitted from the kramapiifha, which therefore merely states: f siinaseieehe-
paf[l nfditam f and then continues with f nfditaf[l sahasriit f etc.
The f!.kpriitisiikhya also mentions an alternative (11.14), which according
to the commentator Uvata results in:
I sunaseit 1 cieehepam 1 sepannfditam 1 etc.
In other words, here the kramapiitha is established from a hypothetical
padapiitha, derived from the saf[lhitiipiitha.
The question arises what should happen if other words precede such a
variant recitation. An example occurs in B.gveda 9.86.42, a chant in praise
of Soma:
Saf[lhitiipiitha: f dvd janii yiitayann an tar iyate narii ea saf[lsam daivyaf[l ea
dhartari I
Padapiifha: I dvd I janii I yiitayan I an tal; I iyate I I ea I
daivyam I ea I dhartari I
'Joining the two kinds of beings [i.e., human and divine], he
enters between human and d.ivine invocation, to support
them'.
Here again the correct word order of the saf[lhitiipiifha could not be de-
rived from a regularly constructed kramapii{ha. This problem is solved as
follows in f!.kpriitiSiikhya 11.15: padiinupzlrvye!Ja sapurva ii tat as tatn
vyavetarrz ea saha vyavdyi ea 'if it is preceded by other words, these are
recited in accordance with word order, and subsequently the divided and
the dividing word together (with the preceding and the following words)'.
Uva(a explains this by specifying that for the above example the preceding
and the following words are I iyate f and I daivyam f. Thus the kramapiitha
becomes:
I dvd janii I janii yiitayan 1 yiitayann anta/:1 I an tar iyate f iyate narii ea
daivyam I daivyaf[l ea I ea dhartari I
1
anuvrtti: see e.g. Renou 1957, s.v.
23
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
The term aniinupurvyasmphitii was used for these rare sequences of the
smphitiipii{ha where the word order is not in accordance with that of the
padapii{ha. Elsewhere the s a ~ h i t i i may be referred to as iinupt1rvyasa1]1hitii.
What is to be retained from these discussions is that the Vedic ritualists
paid great attention to the fixed word order of the two traditional recensions
of\he Vedic text. In the few cases where these two were at variance, special
precautions had to be taken so that in neither version the correct word order
was lost. This leads to a conclusion which may startle those Vedic philolo-
gists who study the Veda in order to arrive at a description of the Vedic
language. For the Vedic ritualists of the Priitisiikhya period the Vedic
language was an object of study only in as far as it was embodied in the
Vedic tradition. Accordingly they regarded the order of words in the Vedic
language as fixed. (We need not here consider another matter, viz., that also
historically speaking it might be said that the word order in Vedic is less free
than it is in classical Sanskrit.) This also applies to the Mlmamsa.
All this illustrates that from the time of the Pratisakhya literature the
Vedic language, apart from its embodiment in the Vedic texts, was hardly
known. The texts were learned by heart, applied in the ritual and handed
down by oral tradition. Moreover, the meaning of mauy Vedic passages
must already at that time have been lost. The ritualists-linguists were inter-
ested, accordingly, in the utterances of the Vedic texts, not in the sentences of
the Vedic language.
This entire development explains an otherwise very curious controversy
which since the Nirukta is known as 'the controversy of Kautsa'. Kautsa
may have been one of the earliest sceptics known, for his main thesis was
that the Veda was without meaning: dnarthakii '"antriib 'the Vedic mantras
are meaningless'. Apart from scepticism, however, other factors may have
been involved. Reno'-! has drawn attention to the fact that one of the
Priitisiikhyas of the Atharvaveda is also attributed to Kautsa (Renou 1960,
68). The Atharvaveda is largely a repository of magical practices. It is
therefore not surprising that those ritualists who treated the mantras as
charms, were hardly in a position to regard them at the same time aslin-
gu.istically meaningful utterances. Katre, on the other hand, regarded Kautsa
as a ritualist who wanted to provide a rational foundation for the ritual
(ibid.; cf. also Thieme 1931, 27).
The Nirukta strongly objected to Kautsa's revolutionary doctrines and
attempted to refute his arguments. One of these is significant in the present
context. Kautsa supported his thesis by reasoning from the premiss that
the Vedic words are fixed and that their order is fixed: niyataviico yuktayo
niyatiinupurvyii (mantriib) bhavanti 'the mantras are joined with fixed sound
and fixed order' (Nirukta 1.15). The implicit argument is that if the mantras
24
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
were meaningful, their words could be replaced by synonyms, i.e., other
words with the same meaning, and the order could be changed at will,
just as in ordinary speech. The Nirukta commentator Durga makes this
quite clear. Consider for example, he says, the mantra :Rgveda 6.16.10
( = Samaveda 1.1): agna d yiihi vitaye '0 Agni, come to the feast'. This is
not the same as : vibhiivaso -lgaccha piiniiya '0 fire god, proceed to the drinking'.
Nor is part of this expression the same as: d yiihy agne. But in ordinary
speech (loke) such conversions are met with. There are many synonyms for
go; moreover, we have govim abhyiija and abhyiija gmJim 'drive the cow
near'; tihara pdtram and pdtram iihara 'fetch the bowl'. This is not so in
mantras. Since we see that they differ from ordinary speech, the mantras
must be meaningless.
This controversy shows that many considered word order in ordinary
speech as free. The author of the Nirukta, however, does not agree with this,
for he replies (Nirukta l.16): apy et ad yathendriigni pitiiputriiv iti
'this [fixed word order occurs] also in expressions of ordinary language, e.g.
indrdgnl "Indra and Agni", pittiputrau "father and son".'
It is not known whether Kautsa had followers who replied to this. We
could easily do so, for both of the Nirukta's counter-examples are nominal
compounds, in which the order of the members is indeed fixed. In indriigni
the order is fixed because in dvandva compounds the member which begins
with a vowel and ends in a short a should precede (this rule is given
by Pal)ini 2.2.33). In pitaputrau the order is fixed because in dvandva
compounds the member which refers to what is more worthy of respect
should precede (this is stated by a viirttika following Pal)ini 2.2.34).
We shall see that these discussions are taken up afresh by the philosophers
of the Mimiiipsa. They also recur in Siiyal).a's commentary on the :Rgveda.s
The Brhaddevatii begins with the statement that its author will list the
Vedic deities as they occur samdmndydnupUrvaSal;z 'in the sequence of the
traditional text' (1.1 ). This expression is adopted from the Brhaddevatii by the
f!.gvidhiina (1.2), where it is used with regard to ritual rules (transl. Gonda,
10). In the granimatical section of the Brhaddevatii some remarks are given
about the grammatical, syntactic and semantic interpretations of the
Vedic text. The only passage which deals exclusively with syntactic matters,
contains a reference to word order (2.100):
atiriklaJ?1 padalfl hinaf/1 viikye nivdayet / ea saf/1-
dadhyiid iinupiirVIf/1 ea kalpayet //
'A redundant word should be rejected, while one that is lacking one should
s See below, page 48.
25
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
introduce into the sentence; and one that is far removed one should bring
into juxtaposition, and should [then] arrange the regular sequence [of the
words]' (Macdonell's translation).
Since the construction of a specific arrangement is here insisted upon, it
seeps that this passage, at least for the language of the Vedas, presupposes a
regular word order. This is explained most satisfactorily by assuming once
more that the ritualists-linguists were interested not in the sentences of the
Vedic language, but in the utterances of the Vedic corpus. We shall study
other supporting evidence for this view.
B. PA1,!INI AND PATANJALI
When entering the domain of vyiikaraQa proper we are no longer concerned
with ritualists interested in preserving the Vedic tradition, but with linguists
studying both the spoken language and, possibly to a more limited
extent, the Vedic language (chandas) (Renou 1967).
A few remarks may first be made about the word orderin Pii1,1ini's meta-
language, i.e. in the itself. Many rules of the grammar prescribe
a substitution. As we have noted elsewhere (Staal 1965a) Pii1,1ini uses the
case terminations in a technical metalinguistic sense for expressing what is
substituted for what, and in what context. Utilizing context-sensitive rules,
substitution of the type Pii1,1ini is interested in may be expressed as:
a[b--+c]d (19)
where c is the substitute for b, wheu preceded by a and followed by d.
Pii1,1ini expresses this by:
a+ Ab!. ending b+ Gen. ending c +Nom. ending d + Loc. ending.
(20)
As in uninflected languages in general, the order of the elements in the
artificial expression (!9) is fixed, since their function depends on their place
in the expression. The order of the elements in (20), on the other hand,
is free, since their function is fully determined by the respective case termi-
nations.
It has been noted that in Pii1,1ini's grammar in general the order of words
is relatively free. Even the order of members of the dvandva compounds
(which yielded the counter-examples provided by the Nirukta) is not fixed
and sometimes deviates from that prescribed by the grammatical rules
themselves. Renou, who has devoted a short study to the word order of
the A$fiidhyiiyi (Renou 1955) states that "l'ordre des mots chez P., d'une
maniere generale, est deroutant. I1 y a des tendances precises, mais rarement
26
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
sans exceptions, et les exceptions sont loin de se reduire a des lois coherentes".
After studying a few special terms (chandasi, nityam) and a few
general cases (e.g., introduction or definition of an anubandha or other
technical term), Renou concluded that "l'ordre des mots chez P., au moins
dans certaines amples categories, comportait une part de Iiberte".
A rather celebrated case deserves a somewhat more detailed treatment; on
this occasion the commentator Pataiijali becomes quite explicit. The first
sutra of the grammar, vrddhir iidaic (I. 1.1 ), defines the technical term
vrddhi. 9 The commentators accept as usual practice in such cases that a
newly defined technical term is put at the end of its definition. This does
indeed obtain in the next rule, aden guQa/z (1.1.2), which introduces the
technical term guQa. The general convention is that the technical 'defining'
term (saf[ljiiii) which is the subject introduced by the siitra and the grammati-
cal subject of the sutra, follows what it denotes, i.e., the saf[ljiiin or 'defined
term' which is the predicate of the siitra ad loc.).
1
0 (Note that
what holds for the object language need not be laid down for the metalan-
guage of grammar: na yathii lake tathii 'it is not so that what
applies to ordinary speech applies to grammar itself' (ed. Kielhorn, I,
39 line 10).)
After suggesting various solutions for this puzzle Patanjali gives as a
final answer: et ad ekam tictiryasya mmigaliirthaf!'l mr$yatdm. 'let the teacher
be forgiven for this once on account of auspiciousness'. From this passage,
the expression maizgaltirtham 'on account of auspiciousness' acquired a
fame. This must be explained as follows. The word vrddhi is not
only a technical term but also a common word of the object language,
meaning 'growth', 'success'. So Pataiijali coutinues: miiftgalika iiciiryo
mahata/z .iiistraughasya maizgaliirthaf[l vrddhiSabdamiidita/z prayuizkte I man-
galiidfni hi .iiistriiQi prathante bhavanty
ctidhyettiras ea vrddhiyuktii yathii syur iti 'the teacher, intent on auspic-
iousness, uses ti1e word vrddhi at the beginning of his great compilation of
science for the sake of auspiciousness. For only such works as have aus-
piciousness at the beginning thrive well and make mankind strong, long-lived
and prosperous' ( ed. Kielhorn, I, 40 lines 6-9).
This discussion makes clear that the Sanskrit grammarians needed a
9
Cardona (1966, 311), whilst largely adopting the methods introduced in Staa11965a, has
criticised the formalisation of the definition of vrddhi (ibid., 66). This formalisation seems
justified, however, in view of 1.1.3 iko gw.wvrddlzi 'gul)a and vrddhi occur in the place of
i, u, r and!' and of 1.3.1 0 yathiisamkhyam anudeSab samiiniim, which specifies (in Cardona's
own terms: p, 309) 'that when items of a subsequent enumeration in a rule of replacement
have the same number as items previously enumerated in the rule, they are related in
order'.
10
See also Scharfe 1961, 21 sq., which however has to be used with caution (Staal 1963b).
27
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
.special convention for the meta-language in order to determine the word
order in a case where ambiguity might result; we shall return to this (below,
page 68). It is not quite clear whether either or both of the following are
ambiguous: what are the subject and predicate of the sentence, or what
are)ts topic and comment (for the distinction cf. Staal 1967a); this is ir-
rel.,Yant in the present context. But it is important to see what Pataiijali has
further to say about this subject in general.
In the course of the discussion Pataiijali has considered another argument,
which does not seem to be taken quite seriously in the grammatical context
though it is stated as valid for the spoken language (cf. again the maxim
just quoted: na yathti lake tathti vytikarm:ze 'it is not so that what applies to
ordinary speech applies to grammar itself'). We need not pay attention to
the word order in the szltra, says Pataiijali, because: neha prayoganiyama
tirabhyate 'restrictions on usage are not here clung to' _11 This he explains as
follows: sal[lskrtya padtiny uts!yyante
bandho bhavati i tad yatha iihara patral[l patram cihareti 'words are generated
in accordance with grammatical rules, but their order [abhism17bandha] is
free, as in iihara pdtram and pdtram iihara "fetch the bowl'" (ed. Kielhorn, I,
39, lines 18-9).
The example we met with in the commentary on the Nirukta is used
again here to illustrate the free order of words in ordinary language."
\ Pataiijali refers to the freedom of word order by using the term yathetam
,. 'according to wish, as you wish'. In simiiE.r contexts this is also referred to
l by such terms as kiimaciira, yathiikiimam 'according to desire'. In this
. )! passage, then, Patanjali clearly regards word order as free. Which particular
order is used in a given utterance is a matter of 'usage' (prayoga).
Before studying additional evidence in this respect from the
we shall return to Piil)ini and see whether he has anything to say about
word order in the Sanskrit object-language. First, it strikes us that the terms
met with before, dnupiirvya and iinupiirvi, are used in a different sense. The
grammarians generally use tinupiirvi, not to refer to word order but to denote
the order in which the grammatical rules are applied (Renou 1957, s.v.).
The order in which the grammatical rules are actually given is in fact much
more important than has even recently been claimed (see Staal 1966c in
11
Subrahmanya Sastri (1960, 145 note 3), erroneously identifying usage with syntax,
concludes that "PiiQ.ini does not deal with Syntax but only with Phonology and Morphol-
ogy". For niyama cf. also Thieme 1931, 30.
12
Note that the examples with the Imperative iihara 'fetch!' cannot be explained from
Patailjali's context, but are natural in the context of the commentary on the Nirukta,
where a Vedic imperative (ii yiihi 'come!') had just been quoted. This suggests that Patai'i-
jali quoted from the Nairuktas, and not vice versa (it is well known that Patai'ijali does
sometimes quote from the Nirukta itself; see, e.g., SkOld 1926, eh. V).
28
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
reply to Fowler !965). PaQini himself (2.1.6) uses iinupiirvya (Boehtlingk:
'ein Nacheinander') differently again, i.e., as a kind of semantic marker
introduced to describe the meaning of a class of first members of indeclinable
(avyayibhiiva) compounds such as 'beginning with the oldest' or
'in order of age (?)' (from 'eldest').
Secondly, PaQini gives many rules for the order of parts of words (e.g.,
stem, suffix) and for the order of members of compounds, which are also
parts of words since nominal compounds are treated as nouns (Staal 1966a,
168-9). He nowhere seems to give any rule for the order of words in the
sentence. But this does not imply, as we shall see, that he gives no rules for
the relation of words in the sentence.
The commentators have noted this and have discussed the apparent
counter-examples. When commenting upon the word order of the meta-
language used in the first siitra, such scholars as Kaiya(a in his Pradipa
(11th century) and Haradatta in his Padamaiijari (not earlier than the 13th
century) statethatPal)ininever discusses the order of independent (svatantra)
words (Ojihara and Renou 1960, 3-4, upon which the following discussion
is partly based). This is shown by refuting the apparent counter-examples,
i.e., PaQini 3.1.2, 2.2.30-31, 1.4.80, 3.1.40-41. The sutra 3.1.2 (paras ea)
lays down that suffixes follow stems, but neither of these are independent
words. 2.2.30 (upasarjanarrz piirvam) and 2.2.31 param) deal
with the order of members of nominal compounds, and the same holds for
2.2.31-38 (not mentioned in this connexion).
The siitra 1.4.80 (te priig dhlitob) says that preverbs and certain other
elements precede verbal roots. This is more interesting, since preverbs,
which by PaQini's time behave as prefixes, could to some extent be considered
independent words since they behaved as such in Vedic. Pal)ini knew this,
and'the next siitra 1.4.81 (chandasi pare'pi) states that in the Veda they may
aiso follow the verbe I root. The commentators provide the examples: for
classical Sanskrit, the former rule e11ables us to derive niydti hastind 'he comes
along with his elephant'; the latter rule provides for Vedic yliti ni hastinii. (This
is not a Vedic quotation: note that the Sanskrit grammarians describe the
Vedic ianguage and are no longer exclusively interested in the Vedic corpus.)
The siltra 1.4.80, then, describes a certain stage in the development of
the language where elements that earlier used to move more freely, have
been given a fixed position and have been attached to other forms. In the
last apparent counter-example the reverse obtains. Here Pal)ini describes a
relation between certain elements in terms of a kind of suffixation, while in
the later development of the language these so-called suffixes started to
move about more freely. Since these suffixes are themselves verbal forms,
Western linguists in their diachronistic tradition describe them as such.
29
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
Pii!)ini on the other hand, finding these elements after verbal forms only,
in his synchronistic manner described them as suffixes. In Western grammars
this grammatical structure is known as the periphrastic perfect.l3 Piii;tini's
sfttras 3.1.40 (krii ciinu pra yujyate liti) and 3.1.41 (vidtirrz kurvantv ity
anygtarasytim) describe the formation for the elements derived from the
veril,al root kr 'do'. Pataiijali added as and bhU "be", which in fact rarely
entered into this formation in the language of the later Vedic period (Renou
1956, 72; cf. also Thieme 1956, 22). While the second siitra introduces a
rather special and optional form, the first yields the general case exemplified
by pacaytirrz caktira 'he did cooking', i.e., 'he cooked'. Patafijali's additional
observations yield pacaytim dsa and pacaytirrz babhiiva 'he was cooking'.
Note that the rule excludes the order caktira pacayam, tisa pacaytim, etc.
That the later commentators considered this rule as a possible, though
apparent, counter-example to their view that Piil)ini nowhere treats word
order, may show that at that time the word character of these forms was
to some extent recognized. That PaJ).ini specified the order (thereby exCluding
cakara pticaytim or the intervention of other words) shows, however, that
he regarded this process as a kind of suffixation. The commentators were
accordingly right in claiming that also in this case Piil).ini was not discussing
the order of words.
If it he then accepted that PiiQini nowhere discussed the order of words,
the question arises whether he considered word order as free. One need not
expect an explicit statement of Pal).ini to thi effect: he is always content to
enumerate rules, mostly abiding by the economy criterion, and hardly
indulges in loose talk. So we shall have to turn to the commentators again,
in this case to the vtirttikaktira, Kiityiiyana. We shall also see, however,
that Piil).ini himself gives many rules about the relations between words in
the sentence. This will clearly show that he was interested in syntactic or
grammatical relations, which are quite independent of word order. At the
same time it will become more apparent that the view that Piil)ini did not
study syntax is at least partly a myth, based upon the assumption that
1
syntax is the study of word order. This assumption results from neglecting
I
; the distinction between the relation of words in a sentence (sarrzbandha),
which certainly belongs to the deep structure, and the order of words of a
sentence (abhism]1bandha), which may belong to the surface structure .
. It has already been noted that Patafijali regarded word order as free. That
this was also the view of Kiityayana follows from an explicit statement in
a vtirttika, which Pataiijali quotes with approval and which he applies in
order to solve a rather complicated Piii;tinian problem.
13 Partly similar observations with regard to the periphrastic future occur in Rocher 1965.
30
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
The background is quite general, the subsequent discussion rather
technicaL When certain forms are substituted for others, the question
arises whether rules which apply to the sounds of the original should also
apply to the sounds of the substitute. The answer is in the negative: Piil)ini
1.1.56 says sthtinivad tideso'nal vidhau 'the substitute [tidda] is like the
original [sthtinin] with reference to a rule, excepting its sounds [a/]'. The
following is an example for the added clause, which is what interests us here.
In 7.1.84 a simple substitution is described: tbe original v in div is replaced by
the substitute au, yielding diau and subsequently dyau. Now 6.1.68 says
that under certain circumstances a final sound disappears: for example,
rtijan looses its final n and is replaced by raja. This would apply to the origi-
nal div in the above situation, yielding di, but not to its substitute dyau
where the conditions of 6.1.68 are not fulfilled. If it were to apply to the
substitute dyau merely because it is a substitute for div, au itself would disap-
pear and we would once more be left with just di. Thus the entire effect of
the substitution would be lost.
After enunciating this very general principle with its qualifying clause,
Piil)ini gives an exception to the qualifying clause in tbe next sutra, 1.1.57:
aca/; parasmin purvavidhau 'the substitute for a vowel is like the original
with reference to a rule regarding a preceding sound'. This, then, is an ex-
ception to the exception expressed by the qualifying clause . .The following is
an example. In pa{aya the rule 7.2.116, applying to the second occurrence
of a, in fact concerns a preceding sound (i.e., the first occurrence of a)
since the result of its application would be pti{aya. Now pti{aya is ungram-
maticaL The form pa{aya itself is derived from pa{u-aya, where u is the
original and zero is the subsitute. It is only the virtual presence of the original
u which prevents the application of7.2.116 and thereby prevents the genera-
tion of the ungrammatical form pti{aya. Summarising: 7.2.116 is a rule
which applies to a preceding sound; the preceding sound is zero, for zero
is substituted for an original u; according to 1.1.57, the rule 7.2.116 should
apply to the original of the preceding sound, i.e., to u; but u does not fulfil
the conditions required for the application of 7.2.116 (the 'analyzability'
condition); hence there is no scope for 7.2.116 to apply, and the ungrammati-
cal form pti{aya is not generated. This is just what is wanted.
The next rule, 1.1.58, enumerates exceptions to this exception to the
exception expressed by the qualifying clause of 1.1.56. Rule 1.1.58 says:
na padiintadvirvacanavareyalopasvarasavarl)iinusviiradlrghajaScarvidi$u. This
means. ~ h t in a number of cases, i.e., padiinta, dvirvacana, vareyalopa,
svara, savarva, anusvdra, dirgha, jaS, and car, the sound substitute is again
unlike the original with reference to a rule regarding a preceding sound.
We shall not analyse and illustrate all these cases but single out dirgha, a
31
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
long vowel substituted for a short original, as in the following example.
In divna the rule 8.2.77 regards the preceding sound i and prescribes that
it be lengthened, with the result that divna is produced. This is the grammatical
form which is required. But divna itself, according to 6.3.134, results from
diva_na, with a as the original and zero as the substitute. However, 8.2.77
not apply to divana, since the conditions for its applicability are not
fulfilled. Hence we do not want the substitute zero to be treated like the
original a, for in that case we would not be able to derive the grammatical
form divna.
What has happened so far? The rule I. I .56 says that there is sthiinivadbhiiva
'[the substitute's] being like the original' in general, but not with regard to
its sounds. The next rule, 1.1.57, says that there is sthiinivadbhiiva also with
reference to sounds, provided the substitution rules regard preceding sounds.
The following rule, 1.1.58, says that there is again no sthtinivadbhiiva in a
. certain number of cases, even though the substitution rules regard preceding
sounds.
At this point Pataiijali, who guides us through such tangles
1
4, says that
with reference to a rule the substitute zero (lopa) in divna is not like the
original (na sthtinivad). This is not, however, formulated in precisely these
terms in I. I .58, and he therefore stages an opponent who objects that it should
have been mentioned explicitly. But no, replies Pataiijali, there is no need.
For all the words which would be required to formulate this explicitly
already occur in this rule and in previous rules. They merely need to be
arranged in such a way that the required statement is obtained. At this
point he quotes in oupport Katyayana's viirttikti which enables us to return
from this long excursus to our proper subject: iinupUrvyel)a
yathe$[am abhisaYflbandha/:t sakyate kartum 'any desired order may be
established between words arranged in a particular succession' ( ed. Kielhorn,
I, 152, lines 24-5; quoted Renou 1957, 57).
The opponent raises a further objection: na caittiny iinupUrvyel]a SalJ.mi-
vi$fiini 'but these are not arranged in a particular succession'. To which the
reply is: antinupi1rvye1Jtipi abhisaJ?1bandho bhavati
'free word order may also be established between words not arranged in a
particular
The term tinupilrvya is here used to refer to a given particular succession
of words; the term abhisGI?Jbandha for word order in general, and the term
yathe${am for 'according to wish' or 'free'. Pataiijali rounds off this discussion
with the following example: anatfvaham udahtiri yti tvaYfl harasi sirasti
kumb/zaYfl bhagini sticinam abhidhtivantam adriik$i/:t may also be arranged as:
14
However, the example for 1.1.57 has been taken from Renou's translation of the A,<>tti-
dhyiiyi; it is based upon the KiiSikii.
32
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
udahari bhagini ya tvmp kumbhal]'l harasi sirasa anatfvahmp sacinam abhidhli-
vantam '0 water-carrying sister, you who carry the pot on your
head, saw the bull running across' (ed. Kielhorn, I, 152line 26 -153line 3).
In order to see what such a re-arrangement amounts to we shall analyse
this sentence in terms of the system ofChomsky's Aspects. The basic sentence
can be represented by the following (already simplified) tree (note that the
order VP-->NP"V has been adopted), (corresponding to the sentence)
bhaginy ana(ivaham adrlikeib '0 sister you saw the bull':
-------
NP
I
N
bhagini
s
VP
/ ',....._
// ....._, (iij)
NP V
I
N
ona(jvflhom
The recursive rule (3) can now be utilized to embed three trees into (iij),
corresponding to the sentences: bhagini Sirasd kumbha111 harati 'the sister
carries the pot on l;l.er head' (S
1
); bhaginy udaka1]1 harati 'the sister carries
water' (S
2
) and: ana(ivahab sdcinam abhidhavati 'the bull runs across' (S
3
).
This results in the following tree (note that the Instrumental sirasli 'upon
the tead' and .the Adverb sacinam'across' have been added without explicitly
formulating the required rules) (see diagram iv, p. 34).
If (iv) is interpreted in the usual way as what has been called (above,
p. 15) a trimmed tree, it would produce the following string: bhagini
bhaginf bhagini udakam harati Sirasti kumbham harati ana{iviiham anarjvdhab.
sacinam abhidhavati adriikeib. Presupposing that we have the required
transformational rules at our disposal, it may be assumed that this can be
transformed into the sentence: bhagini ya tvam udahliri sirasa kumbhmrt
harasi ana(ivahal]'l sacinam abhidhiivantam adriikeib (21). Another arrangement
could be derived by reversing the order of embedding of S
1
and S
2
into (iij).
This would result in the string: bhagini bhagini bhagini sirasii kumbham
harati udakam harati ana(ivaham ana(ivaha/1 slicinam abhidhtivati adrakeib.
This, presumably, could be transformed into the sentence: bhagini yii tval]'l
sirasii kumbhal'/2 harasy udahiiry ana(ivaha1]1 sacinam abhidltiivantam adriikei/1
(22). Still other arrangements could be obtained by changing the order in
some of the re-writing rules utilized to set up (iv).
33
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
s
V
ana<fvlihal) sOcinam abhidhlivoti
bhagini horati
(iv)
N
udakam
Unfortunately, neither (21) nor (22) coincides with either of Pataiijali's
arrangements. Sentences derived with the help of differently ordered re-
writingrules would deviate even more. If one should wish to derive Pataiijali's
sentences from (21) or (22), this could only be done by adopting another
array of transformational rules. The resulting description would make use
of rules like (16)-(18), quoted above (p. 12) in connection with Rama's
looking at Govinda, but it should be clear that the description here required
would turn out to be much more complicated than the one given earlier.
If we look a little more closely, it will become apparentthattheseproblems
cannot be solved even if re-writing rules in terms of sets and wild trees are
adopted. In wild trees, as observed before, the points dominated by a
single node constitute a set. They can accordingly be arranged in any desired
order. But in order to derive Pataiijali's sentences from (iv), whether or
not interpreted as a wild tree, or from any similarly constructed tree, re-
arrangements are required which go beyond the confines of single constitu-
ents or elements dominated by a single node or category symbol.
This may be shown, e.g., by adapting the method of labelled bracketing to
wild trees. It is known that for trimmed trees, an equivalent structure in
terms oflabelled bracketing can always be given, and vice versa (Chomsky &
34
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
Miller, 1963, 288). For wild trees this can be done analogously, using
braces instead of brackets and replacingconcatenation signs (which, however,
are often omitted in labelled bracketing) by commas. The difference between
the two methods may be illustrated by the following example. Application
of the rules:
S -+NP"VP
VP-+V"NP
can be expressed by the trimmed tree:
A
NP )\
V NP
I I
Rfima saw Govindo
or equivalently by labelled bracketing, as follows:
( ( Riima ) ( ( saw ) ( Govinda ) ) ) .
SNP NPVPV VNP NPVPS
Analogously, application of the rules:
S -+{NP, VP}
VP->{V, NP}
(v)
can be expressed by a wild tree which looks exactly like {v), or eqdvalently
by the following labelled and stratified set:
{ { Riima } , { { saw } , { Govinda } } } .
SNP NP VPV V NP NPVPS
Applying this to (iv), interpreted as a wild tree, and omitting innermost
braces, one derives:
{ { bhagini, { { bhagini, { bhagini, { udakam, harati } } } ,
SNP S, NP S
2
VP VP S
2
NP
, { sirasii, kumbham, harati } } } , { { ana<fvaham,
VP VP S
1
NP VP NP.
, { ana<fviihab, { siicinam, abhidhiivati } } } , d r i i k ~ i b } } .
S
3
VP VP S
3
NP VP S
35
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
Assuming again that the various transformational rules have been applied,
one may omit labels, simplify rather liberaiiy and arrive at the foiiowing
stratified set:
{{bhagini, yii, tvam, uddhari, Sirasii, kumbham, harasi}, {anat;lviiham,
\siicinam, abhidhiivantam }, adriik$fb}.
'
Having thus reduced hierarchical complexity, one is left with a set con-
taining three elements, two of which are subsets. If we adhere to the simple
principle that the elements of a set can be arranged at will, it has now become
clear to what extent the order may be re-arranged: the seven words be-
ginning with bhagini may be arranged in any order; the three words beginning
with anatfvtiham may be arranged in any order; and the two subsets them-
selves along with adrtik$i/:l may be arranged in any order. To obtain a
sentence we finaiiy drop ail braces and apply sandhi rules to the resulting
string.
This leads to the conclusion that ev.en if we use sets and wild trees and
aiiow for considerable simplification, we still cannot derive the first of
Pataii.jali's sentences - the simple reason being that anat/vtiham 'buii'
cannot be taken out of its subset and placed elsewhere. One might object to
this long analysis by claiming that the example is far-fetched and that buiis
by nature cannot be kept within narrow confines. But this is not so. Anyone
who wishes to find precise rules for word order in Sanskrit, will constantly
meet with similar problems, provided the sentences he studies are not
uncommonly short.
So it need cause no surprise that Piil)ini did not provide rules for word .
order in Sanskrit. In fact the Sanskrit grammarians assumed, as we have
seen before, not that such ri1!es are beyond the scope of grammar, but that
there are no such rules, since word order or abhisa1]1bandha is free. But if
they neither studied word order nor neglected syntax, what then are the
grammatical relations (sarytbandha) which the Sanskrit grammarians did
study and by means of which they analysed the. construction of sentences?
To answer this problem in fuii a considerable number of Pal)ini's rules
would have to be analysed. Since this seems to be beyond the scope of the
present investigation, which primarily deals with word order, we shaii only
consider the topic which in this respect is fundamental, i.e., the theory of
the relations which Pal)ini caiied karaka. Out of the seven ktiraka relations
which are enumerated, we shaii single out kmman for particular scrutiny.
Even with these restrictions it wiii be possible to show that the problem
of word order is not only affected, but is indeed placed in its proper per-
spective by the ktiraka theory.
The theory of ktiraka has been studied ever since Western scholars
36
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
started dealing with Pat)ini's grammar. The relevant literature includes
contributions by Whitney, Faddegon, and Thieme; the translators of the
grammar, Boehtlingk and Renou, had to face its many intricacies; and most
recently the topic has again been taken up by Rosane Rocher (1964a).
Almost all these scholars, with the possible exception of Faddegon, have
approached the problem in such a way that certain basic ideas have remained
unclear. Still, the questions to which the kriraka theory formulates an answer,
are very pertinent ones. They emerge as soon as one realizes that something
important was missing in the preceding analysis of Patafijali's sentences.
This analysis was inadequate in many respects, but in particular because it
failed to explain why certain declensional and conjugational terminations
occur in certain places. Why did for example bhagini have the Vocative
ending, bhaginz the Nominative, kumbham, anGijvaham and abhidhrivantam
the Accusative, and why were harasi and adrrik$iQ in the Singular? Can such
questions indeed be meaningfully asked after attempts have been made to
establish some order between words that themselves still lack determination?
The karaka theory formulates a solution to precisely such problems.
Though Faddegon's formulations are sometimes weird
1
' and his analysis
of the notion of kiiraka is sketchy and somewhat obscured by the fact that
he treats other problems alongside it, his general poirit of departure (as
quoted above, page 3) was sound and correct. His interpretation of
karaka is formulated as follows: "By kiirakas Plit)ini understands the logical
or ideational relations between a noun and a verb, or more between
an object or anything conceived aftet the analogy of an object and an action
or anything conceived after the analogy of an action" (1936, 18).
In this passage Faddegon seeks to find an adequate expression for certain
grammatical relations of the deep structure. If scholars had by 1936 re-
discovered the importance of deep structure not only forlogic and philosophy,
but also for linguistics, Faddegon would have had an adequate theory and
terminology at his disposal for handling theory. The karaka theory
can in fact be understood from the purely semantic interpretation of the
deep structure. Alternatively and equivalently, it can be undecstood by
studying the various sentences of the surface structure, in which identical
grammatical relations of the deep structure are expressed.
An example may make this clear. Let us consider the kiiraka relation of
/carman, which expressed the direct object. A simple sentence which may be
immediately generated from the base is:
kumbhan karoti 'he makes pots'. (23)
Hi Cf. Renou 1940, 10 note 1: "ouvrage un peu fantasque, mais seduisant".
37
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
Here the kiiraka relation karman obtains between the Verb karoti and the
direct object kumbhiin. This relation could be expressed in Chomskian terms
as [NP, VP]. In the particular sentence (23), it is expressed by the Accusative
ending -an which is attached to the Nominal stem kumbh-.
The kiiraka relations may in general be symbolized by [NP, VP]. They do
nofgenerally include the Subject (in fact there seems to be some inconsistency
here) and accordingly may not be symbolized by [NP, S]. Direct relations
between the denotations of two nouns are also excluded. This is again
interesting in the context of generative grammar, since such relations (as
sometimes manifested, e.g., by the Genitive case) do not belong to base
Phrase-markers but are introduced by embedding Phrase-markers into each
other, i.e., by applying a rule like: NP-->Det"'N"'S.
Since the Subject and what is expressed by the Genitive (but not the
Objective Genitive: see below, page 40) are excluded from the kiiraka
relations, we are left with the relations between the verb and the different
noun phrases of the verb phrase. In terms ofthe tree:
(vi)
we can state that karaka relations are [NP
1
, VP], ... , [NP., VP]. These
relations would of course have to be further determined. But first let us return
to Piil)ini.
Before showing how kiiraka relationships are manifested in basic sen-
tences and, equivalently, in transformationally derived sentences, Pa.JJ-ini
gives semantic characterizations. These could be translated variously; their
precise interpretation and translation continue to raise problems and will
determine the interpretation and translation of later rules. We shall pro-
visionally interpret them as follows, mainly following Faddegon (1936, 18)
and Renou's translation of the The kiiraka relations are denoted
by the following terms: apadtina, Sa!J1pradiina, kara1Ja, adhikaral)a, karman,
kartar, and hetu. The term apadiina denotes the relation of movement
away from a fixed point (1.4.24); sarttpradana denotes the relation with
whom or what one has in view, when giving something (1.4.32); karava
denotes the relation with the most effective means (1.4.42); .adhikarava
denotes the relation with the locus (1.4.45); karman denotes the relation
38
I
I
i
!
i
I
I
I
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
with what is primarily desired by the agent (1.4.49); kartar denotes the
independent (agent) (!.4.54); hetu, finally, denotes what prompts the agent
(!.4.55).
This is made more precise in syntactical terms, when we are told how most
of these relationships are expressed in sentences immediately derived from
a base Phrase-marker: apaddna is expressed by the Ablative (2.3.28) e.g.,
grtimtid dgacchati 'he arrives from the village'; SGf!lpraddna is expressed
by the Dative (2.3.13), e.g., viprdyagdl!1 daddti 'he gives a cow to the Brahman';
karm;a is expressed by the Instrumental (2.3.18: but see below), e.g.,parasunii
chinatti 'he cuts the tree with an axe'l6; adhikaral)a is expressed by
the Locative (2.3.36), e.g., kat a dste 'he sits on a mat'; karman is expressed
by the Accusative (2.3.2), e.g., kumbhdn karoti 'he makes pots'; kartar
is expressed again by the Instrumental (2.3.18: see above and below), e.g.,
vipreQa pacyate 'it is cooked by the Brahman'; he tu, finally, is expressed
as the Subject of a Causative verb (!.4.55), e.g., kdrayati Devadattab 'Deva-
datta causes [someone else] to make'.
In the first five cases we are shown how kdraka relationships are expressed
in sentences immediately derived from a base Phrase-marker; in the two
remaining cases we are shown how they are expressed in transformationally
derived sentences (i.e., Passive and Causative). This apparent exception is
due to the fact that in this system the Nominative occupies a very special
place, which we cannot discuss here (see Thieme !956).1'
Most Western interpreters have been puzzled by the fact. that, in addition
to straightforward cases and their superficial expression by means of case
terminations, PaiJ.ini seems to have required for his kdraka relations
thing vaguely semantic. This is due to the fact that they assumed that
Pa]fini was only interested in surface morphology, for which cases and
case endings seem to be quite sufficient at first sight. But Pa]fini was also
interested in syntax and sentence construction, or in what is nowadays
called deep structure. In order to appreciate how this system works syn-
tactically we shall need to look ai the construction of transformationa!ly
derived sentences. It will then be seen that we cannot establish any result
16
Matilal (1961) has reported on interesting discussions between grammarians and
logicians with regard to the view, that karm:za expresses the cause: since in such sentences
as he sees with his eyes' the kara1;a is expressed by the word eye', this would lead to the
undesirable conclusion that the eye causes the visual object. - Such confusions could
provide the background for subjective idealism. It would however be premature to char-
acterize Indian philosophy in similar terms (Renou-Filliozat 1953,7: .. Les organes des sens
ne sont pas des recepteurs passifs d'excitations, ce sont des 'forces' (indriya) qui agrippent
au passage et parfois vont chercher (c'est le cas de la vue) les objets exterieurs").
17
Moreover, the commentator's views on hetu, adopted here, have been challenged by
Rochei (1964b), mainly because the agent of the Causative verb, too, seems to have been
regarded by Pal)ini as expressing kartar.
39
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
without some such notion as kiiraka. This will be shown in a somewhat
simplified manner, mainly with respect to karman, and without giving a
complete analysis of the many rules which are involved.
As we have seen, Pal)ini (2.3.2) says that in sentences immediately derived
from a base Phrase-marker, the karman relationship is expressed by the
Accilsative:
'
kurnbhiin kmoti 'he makes pots'. (23)
But if we derive the Passive:
(tena) kurnbhii/:z kriyante 'pots are made (by him)' (24)
the same karman is now denoted by the Nominative ending in combination
with a Passive marker attached to the Verb. This. follows from 3.4.79,
which introduces Passive markers, and 3.4.69 and 1.3.13, which state that
Passive markers serve to express kmman. Similarly the Perfect Passive
Participle occurs in:
kumbhiif:z /qtiil;z 'pots (are) made". (25)
Here again the same karman is expressed on account of 3.4.70. But also in
the Objective Genitive of:
kumbhiiniil[l kartii 'maker of pots' (26)
this karman occurs, on account of 2.3.65. It appears once more in the
nominal compound:
kumbhakiira 'pot-maker, potter'. (27)
If anywhere, the term 'transformation' is applicable here: (24)-(27) are
all trarisformationally related to (23). Pal).ini expresses this by sayirtg that
karman is the one kiiraka relationship which is present in each of these
expressions. The expressions arc, moreover, synonymous, since transform-
ations do not change meaning: there is, as the commentators say, sdmarthya
'sameness of meaning' (with certain qualifications, some of them mentioned
below).
In order to establish that according to Pal)ini one relationship occurs
in many different linguistic expressions, rules have to be quoted from different
parts of the grammar. Students of Pal)ini are familiar with this device,
for the order of the rules is based upon other considerations such as /iighava
'simplicity' and anuvrtti 'recurrence' (see e.g. Renou 1957, s.v.; cf. Fowler
. 1965, Staall966c). We have already had examples of this (pages 31-2). Lest
it be thought, however, that we want to present Pal)ini as the discoverer of
transformations by making use of far-fetched or even forced interpretations
made possible by this device, we must show that both modern and ancient
40
I.
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
commentators were aware of such connexions as exist between (23)--(27).
Rosane Rocher, who interprets Piil)ini without any reference to problems
of contemporary linguistics, is as explicit as one could desire: "Pour P.:o.ini,
un fait quelconque- celui, par exemple, qu'un personnage fabrique des pots-
peut s'exprimer par une construction active kumbhtin karoti, une tournure
passive tena kumbhii/:1 kriyante, par un nom d'action accompagne d'un
genitif objectif kumbluiniil]'l kartr, on encore un compose kumbhakiira"
(Rocher .1964a, 52).
Pataiijali provides examples of the same kiiraka relationship, though from
different underlying sentences of the base. This choice of examples slightly
obscures the issue, but may explain that the importance of this passage
(and of similar later passages) has generally been overlooked. Having
stated that the ktiraka relationships are invariant in a variety of linguistic
expressions, an opponent says that then parigavanal]'l kartavyam 'an
inventory [of these expressions] should be made'. Pataiijali replies with a
vtirttika which enumerates various technical terms whose function will be
clear from the examples which follow: tinkrttaddhitasamasaib parisankhya-
nam 'the enumeration is given by means of tin [verbal termination],
krt [a class of primary affixes], taddhita [a class of secondary affixes] and
samtisa [nominal compounds]'. He comments upon this viirttika in the
usual manner, quoting it again, and subsequently provides the examples:
'tili as in kriyate kata{t "the mat is made"; krt as in krtaf:z katal; "the made
mat"; taddhita as in aupagava/:1 "descendant of Upagu" or kapatava/:1
"descendant of Kapatu"; and samfisa as in citragul:z "possessing brindled
cows" or sabalagub "possessing mottled cows"' (ed. Kielhorn, I, 441, lines
19-22).
These examples may be explained as follows. In kriyate katab 'the mat
is made' the tin or verbal termination -te indicates that the karaka relation is
karman; the sentence is therefore synonymous, e.g., with ka{alfl karoti 'he
makes the mat'. In krtab katab 'the made mdt' the krt or primary affix-ta(/:1)
indicates that the karaka relation is also karman; the phrase is therefore
semantically related e.g. to the same sentence, ka{al]'l karoti 'he makes the
mat'. In aupagavab 'descendant of Upagu' the taddhita or secondary affix,
technically called av, indicates that the ktiraka relation is apadtina; the
phrase is therefore synonymous with upagor apatyam 'descendant of Upagu'.
Similarly for ktipa{ava/:1. In citragub 'possessing brindled cows', which is a
samtisa or nominal compound of the bahuvrihi variety, the kiiraka relation
is karman; the phrase is therefore synonymous with citragoman 'possessing
brindled cows' or yasya citragtivab.sab '(he) who possesses brindled cows'.
Similarly for saba/agu/:1.
We see from these examples that the Indian grammarians not only
41
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
considered transformational relations, but a great variety of such relations.
In some of these the contemporary maxim 'transformations do not change
meaning' (cf. Katz and Postal1964; Staal 1965b) is adhered to. In others,
closely related semantic relations are envisaged. These various semantic
relations are carefully distinguished (a list of some of these occurs in Renou
195t s. v. 'taddhita') and it may well be asked whether contemporary trans-
formational theory will not have to consider such relations in due course.
In the passage commenting upon siitra 2.3.2 in Bhattoji Siddhan-
takaumudi, a rather similar discussion occurs: anukte karmm:zi dvitiyd sytit /
harilfl bhajati I abhihite tu karma!Ji prdtipadikdrthamdtra iti prathamaiva I
abhidhdnalfl tu prdya!Ja tilikrttadhitasamdsal; I tin haril; sevyate I krt
sevital; 1 taddhital; satena kritaf.z satyal; 1 samdsal; prdpta dnando ya'l' sa
prdptdnandal; 'If the karman relationship is not expressed otherwise it is
expressed by the Accusative, as in harii?Z bhajati "he worships Hari". If,
however, it is expressed by the Verb, itisin the Nominative. The meaning is
generally introduced by either tili, krt, taddhita or samdsa: tifz as in hari
sevyate "Hari is served"; krt as in lak$n'lyci sevita[l "served by Lak$mi";
taddhita as in satya from satena kritaiz "bought with a hundred"; and samdsa
as in prdptdnandal; "blissful" from prdpta tinando ya111 saiz "he whom bliss
has reached".'
The examples here given by Bhattoji may be explained along similar lines,
though is at least one significant extension. In hari sevyate 'Hari is
served' the tbi or verbal ending -te again indicates the karman relation;
but if it is true that this sentence is considered synonymous with harirrz
bhajati 'he worships Hari', as indicated, this is particularly interesting, since
it establishes a semantic relationship between verbs, which are related in the
manner in which in English please and like, buy and sell, etc., are related
(for mechanisms and devices which might deal with such relations, see
Staal 1967a). In sevitaf.z 'served by the krt or primary
affix -ta(/;1) indicates that the kdraka relation is kartar; this phrase is therefore
semantically !elated to sevate serves' (notice that this
analysis is in fact determined by the Instrumental ending, which, however, is a
sup or nominal termination, not a krt; if this be rejected, the interpretation
given to Pataiijali's krtal; katah may be resorted to). In satya the taddhita
or secondary affix -ya indicates that the kdraka relation is kara!Ja; the phrase
is therefore synonymous with satena krital; 'bought with a hundred'. In
prdptdnandal; 'blissful', lastly, the kdraka relation is karman again and the
synonymous expression is given.
These passages may be sufficient to demonstrate that the Sanskrit gram-
marians needed the kdraka theory for establishing relations that we would
nowadays call 'tranformational'. The related expressions may be sentences
42
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
or noun phrases. In the former case nothing is said with regard to word
order; in the latter case the order of the elements may be fixed, as in the case
of the taddhita suffixes. In such cases as krt, the order of noun and verb
is free, but the order of stem and suffix is determined. The case of nominal
compounds is especially interesting, since the fixation of order of the mem-
'
bers is postponed to a later stage than we might expect. For it seems reason-
able to assume,. that a description in terms of strings and based upon con-
siderations of order, though perhaps inadequate for sentences which may
be ordered 'freely', is particularly appropriate for nominal compounds.
This was also more or less assumed in the cases given by the Nirukta as
counter-examples (above, page 25). However, as we have shown elsewhere
(Staal 1966a, 171-2), PaQ.ini first introduces the distinction between main
(pradhlina) and subordinate (upasarjana) members of compounds, which is
determined by their grammatical relationship only. Order is introduced much
leter. In other words, compounds, though ultimately described as strings
of elements, are primarily accounted for in terms of relationships between
these elements which need not be manifest in or derivable from the order of
the elements or their physical shape. This is in fact the main characteristic
of all these relationships. The principle underlying this practice with regard
to the analysis of nominal composition, is succinctly expressed in the
following viirttikii: etad gantavya/[1 ya/[1 prati yad apradhiina/[1
ta/[1 prati tadupasarjana/[1 bhavati 'from the grammatical relation [sm11bandha]
it has to be inferred that a word in relation to which a term is of secondary
importance is that in relation to which the term subordinate (upasarjana)
is used' (ed. Kielhorn, I, 215 lines 10-1, quoted Renou 1957, 326).
The kiiraka theory was not merely adopted as a general theory to be tested
by carefully selected examples: Its application to questions of detail raised
by problems of grammatical description, was also carefully studied. At the
same time the general background and significance of the theory were the
object of research and speculation. An example will now be given of each
of these two approaches before we conclude this section with a general
description of these relations-hips. Both examples are taken from Patafijali's
long discussion following siitra 2.3.1.
The question is considered what should happen if more than one word
seems to occupy the same place in a kiiraka relationship. In such sentences
as kafaJF karoti bhi$mam 'he makes a dreadful mat', ka{al!' karoty udiiram
'he makes an excellent mat', ka{a/[1 karoti .fobhanam 'he makes a beautiful
mat'. ka{af!1. karoti dar.Saniyam 'he makes a handsome mat', the rule explains
why kat am 'mat' has the Accusative ending. But what about bhi$mam, etc.?
Patafijali discusses various suggestions, whicli will be merely mentioned
here. The Genitive, which often takes care of the remaining relationships,
43
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
might in some sense be used. Or else the rule which introduces the Accusative
for katam is somehow construed as also applying to bhi$mam, etc. Or
lastly, bhi$mam etc. follow katam, 'since they occupy the same position'
(siimiiniidhikarm:zytit) (ed. Kielhorn, I, 441line 23- 442line 5).
11sewhere the relation between ktiraka and the ontological categories is
briefly touched upon.'B Nowhere in this connection does Patanjali evince
the philosophical obtuseness of which he has recently been accused (Frau-
wallner 1960, 107). His statement, on the contrary, is forceful and clear.
In addition it is linguistically interesting because it takes into account the
fact that sentences occur in a wider context of discourse.
The idea is entertained that ktiraka relations perhaps always refer to the
same categories, e.g., dravya 'substance'. But it is easily discarded by just
listening to the following conversation: kva devadatta/:z 'Where is Devadatta?'
- asau vrk$e 'On that tree'. - katarasmin 'On which of the two?' - yti t;$fhati
which stands'. This receives the COJ!lment: sa vrk$0' dhikara1JG1!1 bhUtvti
anyena .iabdentibhisal[lbadhyamiina/:z kartti sal[lpadyate 'that tree which was
adhikaralJa becomes kartar as soon as it is combined with another word'
(ed. Kielhorn, I, 4421ines 24-5).
The ktiraka relationships, then, are semantic, not ontological relationships.
In semantical terms this means, moreover, that the karaka relations deal
with meaning, not with reference. They could be called 'logical' in the sense
in which Chomsky used the term when speaking about logical subject,
logical object, eic. (1965, 70). In the familiar tree:
s
NP
A
(vij)
NP V
we may define [NP, VP] as the grammatical Object-of. But in a sentence
which is transformationally derived from (vij), the NP determined by [NP,
VP] may occupy the position of the grammatical Subject. It continues how-
1s Attention may be drawn to the fact that the ontological categories of the
system, which were adopted by the Indian logicians, were based upon categories of lan-
guage, as were Aristotle's categories (cf."Staal1963a, 18-21 =1965d, 106). Faddegon,
who most probably was the first Western scholar to point this out, constructed a 'sema-
siological classification' to explain the relation between Noun and Verb, which at the same
time was to serve as a basis for the discussion of the categories (Faddegon 1918,
108-11).
44
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
ever to be the logical object provided this is defined as the grammatical
object of the underlying sentence. By means of simple definitions of this type
one obtains exactly those relations which Pii]fini called karaka. It need
scarcely be added, finally, that such definitions can be given in terms of
wild trees and, following Pii]fini, without taking order into account.
C. JAIMINI, SAHARA, AND KUMARILA BHATTA
The philosophers of the Mlmiimsii have been correctly characterized as
vdkyamimiirrzsakas 'investigators of the sentence' (Renou, quoted above,
pages 1 and 20). Still, there is little in their teaching with respect to word
order that would not by now be familiar to the reader. They mainly revert
to discussions of the Kautsa-controversy and their entire approach is very
similar to that which was found in the Nirukta (above, p. 24-5). Some of
their principal statements pertinent to our context will now be referred to,
first, in order to show the continuity of these discussions for more than a
millennium, secondly because there are clear indications that some results
of the grammarians have been taken into account. An additional reason is
that in the course of these discussions a few points are stated rather clearly
and an important new distinction is introduced.
In Jaimini's MimiirrtsdsUtra and in Sahara's commentary. we meet again
with opponents who maintain Kautsa's thesis that the Vedic mantras are
meaningless. Many reasons are adduced: so often the mantras appear
simply absurd; they speak of things. that do not exist (e.g., something with
four horns, three feet, two heads and seven hands); they address inanimate
objects (e.g.,herbs and stones); they are self-contradictory and often redun-
dant; there is a tradition for them to be learnt by heart, but no corresponding
teaching of their meaning; etc. (MirntiiJ1stisiitra 1.2.4, sutras 34-8; cf. Renou
1960, 68 sq.). Again the familiar reason is given: vtikyaniyarntit 'because the
order of words (in mantras) is fixed' (1.2.4, siitra 32).
Sabara describes the opponents' view in the following terms: niyatapada-
krarnti hi mantrti bhavanti I agnir miirdhti diva iti na I yady
arthapratytiyantirthti viparyayevapy arthal; pratiyata iti niyamo'narthaka/;l
sydt f viparyaye'nyad ucciiral)am iti niyama dSriyate
tena yatarasmin niyamo'rthavdn sa niinarrz pak$a iti j nanv arthavatsv
api niyamo drsyate I yathendriigni iti I yuktaf/1 tatra tat I viparyaye'rthapra-
tyaytibhiivat 'For mantras have the order [krama] of words fixed [niyata].
For example, agnir m.Urdhd diva, and not the reverse. If meanings were
meant to be expressed, the meaning would also be expressed by the reverse
order, and so fixing the order would be pointless. If it be argued that a
particular utterance [ucctirava] is intended and that that utterance, when
45
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
reversed, would be another one, so that fixed order has for that reason been
established - then, on the second of the views mentioned, fixing the order
would indeed be significant. Further, with regard to meaningful expressions
fixed order is also observed, for example in indriigni "Indra and Agni".
In, this case it is correct, for the reversal would not express any meaning'
(ed, Anandii.srama, 146).
The most interesting observation here made by the opponent is that
word order is fixed in utterances, even if it be not fixed in sentences. And
this much is granted by the author. The distinction is indeed relevant for
the Mimii.msii., which studied, as we have seen, the Vedic language in as
far as it was embodied in the Vedic tradition {above, p. 24). By the time
of Sahara the distinction between sentence and utterance is known (it
could be most easily derived from the distinction between type and token,
cf. Staal 1966d), and Sahara realises that his investigations are concerned
with utterances, studied as such and not as utterance tokens of sentence
types. This might perhaps be interpreted as saying, that the Mimii.msii. is
interested in language from the point of view of performance, but not of
competence {Chomsky 1964, Ch. I, and cf. Levin 1965); in fact !Viimii.msii.
pays no attention to the deep structure which represents competence and
which was studied so intensively by the grammarians. It is clear, at any rate,
that the view according to which the word order of utterances is fixed, is not
primarily relevant to a linguistic investigation into the properties of sentences.
Kumii.rila Bhatta in the Tantraviirttikii gives better examples when
commenting upon this passage and also introduces an additional distinction
between different kinds of semantic anomalies: agnir miirdhii iti yo'rtha/J
pratiyate sa miirdhiignir ity aneniipity anarthako niyama/J 'the meaning of
agnir miirdha is the same as that of miirdha agnil;, hence fixing the order is
pointless'. However, the determination of order is meaningful 'in the case
of stems, affixes and compounds' (prakrtipratyayasamtise$U), for example
indriigni 'lndra and Agni', riijapuru$a 'king's man' and ni$kauSiimbifJ 'one
who has left Kausii.mbi'. He continues: u k t a ~ tatra viparyaye'pasabddrthd-
nyatvdnarthakyaprasaflgiit 'here it is reasonable, since, if the order is
reversed, an incorrect form [apaabda], a different meaning [arthiinyatva]
or no meaning at all [iinarthakya] would result'. Examples follow: because
of Pii.l)ini 2.2.33 (see above, page 25) agnindriiv ity asiidhutvarn 'agnindrau
is not well-formed'. Furthermore, punqarii.ja ity arthiinyatvarrz kauSdmbinir
ity anarthakatvam 'purU$Grii.ja would have a different meaning [i.e., "man's
king"] and kaustimbinil; no meaning at all'. (Note that a .difference is made
between the first and the third form, though both are ungrammatical; it is
felt, apparently, that to the first, but not to the last, some meaning could be
ascribed.)
46
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
These are the Mimal]'lsa renderings and first reactions with regard to the
unorthodox views of the followers of Kautsa. The final answer to the
fixed-order argument is given in a later siitra of the Mimarrzsasiitra (1.2.4,
siitra 44). It is a little disappointing, but quite commonsensical: aviruddharrz
param 'the opponent's view is not inconsistent [i.e., with our own view
that mantras are meaningful]'. Sabara explains this further: kiimam anartha-
ko niyamo na apramtil)am 'perhaps the fixed order is pointless; at
any rate what is observed [i.e., that the words do convey meaning in whatever
order they occur) cannot be rejected' (ed. Anandasrama, 155).
Sahara also hints that the mantras with their fixed order may establish the
unseen result which according to the Mlmal]'lsa philosophers is the fruit of
correctly performed ritual activities. Kumarila Bhatta developes this further:
satsv apy upiiytintarani-
vrttau niyatiidr.Jfamtitrakalpanayii siddham arthtibhidhtinam 'even if there are
other means [to express the same meaning], only what possesses a particular
word order is recognised as a mantra. When the other
means are discarded the fixed unseen result may be reached. Therefore it
is established that mantras convey meaning'.
Later in his work Kumarila returns to the same topic (1.3.9). He reports
a controversy between an opponent, who claims that the Vedic language is
different from ordinary (laukika) language, and the author himself, who
considers both languages as identical. The opponent enumerates points of
difference, including the fact that 'words and sentences' (padaviikya) have
a fixed order (niyata) in the Veda but a free order (aniyata) in ordinary
language. Kumarila argues against this that the two are identical, among
other reasons, because sentences are only difierently but the
words and sounds are not different ( viikyatatsamt!hamiitranibandhanapanna-
( ed. Anandasrama, 291 ).
These discussions show that, even if there is some uncertainty with regard
to word order in the utterances of the Veda, the Mlmal]'lsa philosophers
agree unanimously that word order in ordinary language is free. The dis-
tinction between the fixed word order of Vedic utterances and the free word
order of the spoken language is most probably related to a much broader
distinction between the general fixedness ofVedic utterances and the freedom
of ordinary speech. This view is formulated for example in the following
quotation, attributed to Vardhamana and found in the chapter on gram-
matical philosophy of a 14th-century survey of philosophical doctrines:
janab
tu viSe,Joktif:z pravartatiim
(Sarvadarsanasarrzgraha 1951, 291)
47
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
'In ordinary speech men may proceed according to wish,
But in Vedic paths precise enunciation must be employed.'
{cf. transl. Cowell, 205).
The later speculations ofthe Mlm1up.sa philosophers and their controversies
withl other schools (Buddhists, grammarians, logicians) display a variety of
'
viewpoints and doctrines with regard to sentences and their meaning. These
will not, however, be referred to here (cf. e.g. Staal1967b). Though this
literature is now becoming better known and understood, it would still be
premature to state that it no>vhere deals with word order. From what we
know at present, however, it seems unlikely that any extensive discussions of
this topic occur. Syntactic relations or mutual expectancy (tikditkd) are
often referred to, but they are always independent of word order; they are
concerned with sal]1bandha. 'Proximity' (sal]1nidhi), which is another con-
dition for sentencehood, does not refer to word order but to the requirement
that words, if they are to convey any meaning, should not be uttered at
long intervals (e.g. Kunjunni Raja 1963, 166-9). According to the Miina-
meyodaya, a late work on Mlmarp.sii. epistemology, the rival Mimarp.sii.
philosopher Prabhiikara Guru did not even understand this requirement,
which should be interpreted as referring to sounds or words (sabdasal]1nidhi);
he merely considered 'prox!mity of understanding' (buddhisal]1nidhi) (ed.
Kunhan Raja &Suryanarayana Sastri, 100).
There were many discussions, finally, regar<iing the order of sminds in
an utterance, as well as speculations about the question how sentences are
understood as a unity after the phonemes have been heard in linear sequence.
But this again is a different topic from word order.
That the Kautsa controversy continued to be regarded as a topic worthy
of discussion even as late as the 14th century, appears from Sayal}a's
commentary on the Rgveda (see above, p. 25). Since Sayal}a's treatment is
again similar to that of the Nirukta and of the philosophers of the Mimaq1sii.,
it may be sufficient to refer to Oertel's work for a translation of the text
(Oerte11930, 58-60). Sii.yal}a represents the opponent as saying that mantras
are recited in a fixed order, though their meaning could also be expressed
by a different arrangement of the words. However, it is not the meaning of
mantras that counts, but their correct recitation leading to the required
results. Sii.yal}a states in reply, that he nowhere disagrees with this con-
tention. He merely claims that mantras, in addition to having certain effects,
also have meaning_!
1
9
The passages discussed in this section from the Mahiibhii.yya and from MimS.q1sa do
not throw any light on Frauwallner's suggestion of a later date for Pataii.jali (Frauwallner
1960). From the facts dealt with here it may be inferred that Pataii.jali, who certainly
48
INDIAN THEORISTS ON WORD ORDER IN SANSKRIT
Summarizing, we may conclude that the Indian theorists agreed almost
unanimously that word order in Sanskrit is free. That the Vedic ritualists
regarded the order of words in specified utterances from the Vedic corpus as
fixed is not very surprising, since it is due to the fact that their interest was
mainly textual and, in a sense, philological. This could at most lead to a
certain limited development of phonetics, and this indeed has taken place.
No general linguistic theory can, however, be evolved when utterances are
adhered to and theoretical construction, which leads away from utterances
to the constitution of sentences, is repudiated. The study of grammar could
not develop exclusively from the limited assumptions regarding the nature of
language which characterize the speculation of !he Vedic ritualists and
Mlmarpsa philosophers. For linguistics to originate, a new impetus was.
needed. This impulse was provided by the grammatical tradition, which
stressed the importance not only of common usage (loka) but also of the
infinity of the expressions of language. This new development was carried
to perfection by Pal)ini and his followers.
The difference between the two approaches may be once more illustrated
with the help of a quotation from Patafijali's The issue which is
being discussed is the eternality of the Veda, accepted of course by Patafijali
himself. "Is it not said", he says, "that the Vedas were not composed,
but are eternal? -Quite so, but it is their sense that is so, not the order of the
syllables in them."'"
knew the Nirukta (see above, note 12), was not familiar with the development of the
MimafJ1sasfitra; for whilst he was both intere-sted and versed in ritual matters (see e.g. Puri
1957, 166-79), he does not show familiarity with the Mima:rpsa discussion on word order
in the context of the controvers:y of Kautsa (see also page 49 and note 20). Since
Jaimini's date is quite uncertain (estimates vary between the 2nd century B.C. and the
2nd century A.D.) this does not lead to any definite conclusion. The same can be said
with regard to Frauwallner's discovery of an early Mimarpsa text in the Mahabhti$ya. -
Frauwallner's low assessment of Patafijali's scholarship and place in Indian thot!ght (not
dissimilar to the views defended in Scharfe 1961, but SAC my review 1963b) seems to
reflect a philosophic bias that overrates the importance of epistemological issues and
underrates the relevance of language for philosophy. It is surely preferable to abide by
such views as expressed Thieme 1935b, 181 (" ... die Er6rterung ... zeigt die ganze
Kunst patanjaleischer Darstellung. In abstraktester Form werden die Gesichtspunkte fi.ir
und wider gegeneinander ausgespielt, werden uns die verschiedenen Dberlegungen in
ihren Resultaten vorgeftihrt. Die Ausdrucksweise ist knapp genug, Missverstandnisse des
Interpreten entschuldbar zu machen, und doch wieder so genau, dass die falsche Auf-
fassung eines Arguments sich unweigerlich beim nachsten verrat und racht"), or Renou
1940,22 ("Le texte est Ccrit en prose simple, d'aspect assez archaique, concise et d'une
grande rigueur. Le raisonnement est d'une admirable pertinence ... la dialectique de
Patafijali nous permet s"eule d'arriver au coeur de la pensee piil)inCenne").
20
nanu coktam na hi cchandiimsi kriyante nityiini cchandiimsfti f yady apy artho nityo yii
tv asau van:ziinupiirvi sii'nityii (ed. Kielhorn 2.315, lines 13-4; also quoted Hiriyanna
1932, 313).
49
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
We have at the same time seen that the Indian grammarians paid attention
not only to sentences, but also to sentence construction, to grammatical
relations and to what is nowadays referred to as deep structure. Word order
was clearly excluded from deep structure. In this connection transformational
rel\ltions were discovered, and much attention was given to the semantic
rellltions between transformationally related sentences.
50
CHAPTER Ill
WESTERN SANSKRITISTS ON WORD ORDER
IN SANSKRIT
Western descriptions of word order in Sanskrit have, especially in the
beginning, presented a picture which is fundamentally different from that
given by tbe Indian theorists. These descriptions, which are often confined
to surface structure, and always based upon study of a specified textual
corpus, are characterized by the view, adhered to almost universally, that
there exists a regular or traditional word order and that deviations from
it can be explained. This view, already defended by Th. Benfey and A.
Bergaigne, finds full expression in Delbrueck 1878, a monograph based upon
the study of the SatapathabrdlunaiJa, a large corpus and one of the oldest
examples of Sanskrit prose. A prose text is deliberately chosen since it is
taken to be more appropriate for an investigation into word order in ordinary
SJ?eech, uninfluenced as it is by the exigencies of metre and versification.
Delbrneck arrived at two conclusions: (1) there is a traditional word order in
Sanskrit; (2) exceptions to this order are due to the fact, "dass ein Wort dem
Anfang eines Satzes zurtickt oder an den Anfang rtickt, so bald ein Nachdrnck
des Sinnes auf ihm liegt" (1878, 76):
This 'traditional' word order is described as follows. The subject occurs
at the beginning; the Verb at the end; Dative and Accusative in between,
but the Accusativejmmediately precedes the Verb. Adjectives and Genitives
precede Substantives, Participles and Appositions follow Substantives and
Prepositions follow Cases (i.e., are Postpositions).
This could be described in a system similar to Chomsky 1965, with the
help of the following rules:
S--+ NP"VP (28)
VP--+ (Prep-Phrase )"(Prep-Phrase)" ... "(NP)"V 1 (29)
Prep-Phrase--+ NP"(Prep) (30)
NP--+(S)"N. (31)
This may be symbolized by the following tree:
1
The 'Prep-Phrase' Constituents may also be used to introduce Noun-Phrases with the
termination of Dative, Instrumental, Ablative, and Locative.
51
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
s
---
NP
/\
S N
~
)\ Prep
)\
S N
(viii)
S N
For Participles and Appositions the following transformational rule will
be required:
)\
NP
~
(32)
s N N S
For nominal sentences, i.e., sentences where the Verb is deleted, the
Subject and the Predicate change places (1878, 26-8). In other words, a
transformation basically of the form:
(33)
NP,
is assumed (alternatively, two transformations may be assumed, one for the
deletion of V and another for the reversal of order of the two Noun phrases).
Delbrueck devoted an interesting discussion to the question whether
perhaps conclusion (2) mentioned above would suffice for the description
of word order in Sanskrit: "Man konnte zwar gegen diese Beobachtung
einen Einwand er he ben und behaupten: die Wortstellung war vollkommen
frei, und lediglich diktiert durch das Gesetz,.dass das sHirker betonte Satzglied
vorn steht; die Inder betonten eben das Subject besonders stark, darum
eroffnet dieses stets den Satz u.s.w." (1878, 76). Delbrueck rejects this view
52
WESTERN SANSKRITISTS ON WORD ORDER IN SANSKRIT
on account of two considerations: first, he doubts "ob sich irgendwo im
Bereiche der Erfahrung ein solcher Sprachzustand [i.e., free word order]
findet"; secondly: "Ware die Ordnung der Wi.irtervollkommen frei gewesen,
so miisste sich eine gri.issere Mannigfaltigkeit zeigen, als thatsiichlich
vorhanden ist" (ibid., 76-7). The latter argument seems to be motivated by
impressions based upon frequency of occurrence of certain sequences, i.e.,
upon impressions which are essentially of a statistic nature.
Delbrueck's researches, given wider currency in Delbrueck 1888, provided
the framework for the subsequent elucidation of the problem. Speyer (1896)
adhered to Delbrueck's view and illustrated it with an example of a Sanskrit
sentence taken from the Mahabhd-ya (Speyer 1896, 247): pramd(zabhiita
iiciiryo darbhapavitrapii1Jib Suciiv avakd.e priiilmukha upavi.ya mahatii yatnena
siitrarrz praf)ayati sma 'the competent teacher, having sat down facing East
with purifying darbha grass in his hand in a pure place, composed the rules
with great care' (ed. Kie!hdrn I, 39 lines 10-1).
2
This can be analysed with
the help of the rules (28)-(31), adding:
V-> B"Tns. . (34)
The required analysis may be symbolized by the (simplified) tree on
page 54 (diagram ix).
Here S is: iiciiryo yatnena siitraf{1 prm:zayati sma 'the teacher composed the
rules with care'; S
1
is: iiciiryo 'vakdSa upaviSyati 'the teacher sits down in a
place'; s2 is: dcdryaf:t 'the teacher is competent'; s3 is:
dcdryo darbhapavitrapd(li/1 'the teacher has purifying darbha grass in his
hand'; S
4
is: dciiryal;z prii1imukhab 'the teacht:r faces. East'; S
5
is: avakQSaf:t.
Suci(z 'the"pla:::e is pure'; and S
6
is:yatno mahtin 'the care is great'. Note that
we have not take'n into account that the nominal compounds pramd(labhutab,
darbhapavitrapd(li/;z and prdnmukhafz are also derived t,ansformationally, so
that the branches concerned are in fact longer.
In this way the following string could be derived: dcdryab prdlinzukhab
iiciiryaf:t darbhapavitrapdlJib dcdryab dciirya(z avakd.a(J. .ucil;z
avakiiSe upaviSyati dcdryab yatnaf:t mahdn yatnena siitt'af!J. sma.
Before deriving this, however, the transformational rules (32) and (33) have
to be applied, yielding: dcdryab dcdrya/;1 dcdrya/; darbhapa-
vitrapd!Jil:z dcdryal.z prtiJimukha/J. dcdryal;. avaktiSe Sucib avaktiSal;z upaviSyciti
yatnena mahdn yatnal;z siitram pra7Jayati sma (note that the modifiers recur-
sively introduced by S have been interpreted as Appositions). Applying
2
This passage in the elucidates the maxim: na yathii"loke tathii vyiikaral)e
'it is not so that what applies to ordinary speech applies to grammar itself'(above,page 28).
Scharfe (1961, 10) inferred from this a magical background to Indian linguistics; but see
Staal 1963b, 256.
53
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
s
)\
A 1\
l\ A l\ :::
s, N Prep-Phrase V s, N
1\iiciiryal] \ upavisyjl\ yatnena
l\ V\
S, . N NP S, N (ix)
1'\ciiryaf:l pramiiQabhuta(l f\\vakaie
)\ V\ T
5
4
N NP NP
(\iiciiryaly darbhapavitrapriQi/1 iuci/1
NP VP
acaryaly I
NP
prO.imukhaly
deletion and other transformations (defined in such a way that one occurrence
of actiryab will occupy the desired place), this results in: pramavabhiitab
tictiryab darbhapavitraptivib prtinmukha/:z sucau avaktise upavisya mahata
yatnena sfitram prmJayati sma. If sandhi rules are now applied, the result is
almost the original sentence; prtinmukha/:z is however in the wrong place. We
cannot get this right, for example, by changing the order of embedding of
any of the sentences 8
1
, . , S
4
, as the reader may verify. The results are, at
the most, almost correct. And so we are forced to conclude that Speyer's
example is incorrect, despite the fact that every Sanskritist will immediately
recognize how the sentence has to be interpreted semantically. This merely
confirms our conclusion (above, page 36), that anyone who wishes to find
54
WESTERN SANSKRITISTS ON WORD ORDER IN SANSKRIT
precise rules for word order in Sanskrit will run into difficulties, provided the
sentences he studies are not uncommonly short.
In 249 of his syntax, Speyer dealt with the position of certain particles
which determine the entire sentence. Denoting such particles by II, one
could adopt instead of (28):
S--+II"NP"VP. (28')
These particles are of two kinds: "Die den ganzeu Satz bestimmeuden
Partikeln stehen teils an der Spitze des Satzes, teils werden sie dem ersten
Worte augehiingt." If we interpret 'das erste Wort' as the first NP, Speyer's
rules may be expressed by (28') and:
IT -7 II
1
, II2
II
2
"NP=o- NP"II
1
II
1
--+ ea, vd, u, ha, svid, vai, khalu, kila, tu, na, ...
II
2
--+ atha, uta, api, param, kim, ...
The particle na, lastly, occurs "auch wohl unmittelbar vor das Verbum".
Following the development of the Western study of Sanskrit syntax, we
mention Bloch 1906 only to observe that in this monograph word order
is nowhero discussed. Minard (1936) on the other hand, studying certain
subordinate clauses, again from the Satapathabriihmava,_ described the
regular succession of subordinate and main clause in the compound sentence
which he calls 'dyptique', in almost lyrical terms: "Nulle fantaisie, d'ailleurs,
nulle gr.ce. Partout ... rigueur, un eclat sans dt!faut comme sans sourire, hi
durete attiiante d'un mecanisme de precision. Et d'abord, des deux membres
de la phrase, des deux volets du- dyptique, la sequence meme. est reg!ee.
Jamais elle ne s'inverse, qu'on. n'en puisSe trouver l'origine et la valeur a
du systeme ou dans ses environs immediats, dans la demarche de
la pensee ou dans la pression du contexte" (Minard 1936, 3). Since it is the
subordinate chuse which always precedes, this eau be regarded as a confir-
mation, at least for one type of sentence from the Satapathabriihmava, of
the order established by rule (31).
Canedo 1937 is a monograph especially devoted to the study of word
order. An earlier similar monograph (Thommen 1903), largely based upon
selected passages of the Sanskrit epic, is dismissed with the words: "Die
Wortstellung des M ahiibhiirata, die er besonders untersucht hatte, ist
vo!lstiindig frei" (Canedo 1937, 5; Gouda 1952, however, was to call
this a 'conviction personelle', not a 'demonstration'). Canedo claimed that
he studied grammatical regularities, and did not deal with what would
nowadays be relegated to performance: "Die Wichtigkeit der psychologischen
Momente fiir die Wortstellung soli selbstverstiiudlich nicht in Abrede gestellt
55
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
werden; aber die 'Erkiarung' gehort eher zur Psychologie des Stiles" (6).
Canedo objected to Delbrueck's distinction between traditional and 'oc-
casional' word order, but mainly on historical grounds: the so-called tra-
ditional word order may be recent, while the so-called occasional word
order is often ancient. Speaking about the position of the Verb, Canedo
s a i ~ "Durch alle Perioden der Sprache ist die Endstellung am hiiufigsten
(habituell), hiiufiger als die beiden anderen zusammen, und die Anfangs-
stellung ist hiiufiger als die Mittelstellung" (27).
Adopting (14) and (15) as initial rules (above, page 12), the position of
the Verb which comes next in frequency could be derived by applying (16)
and (17). The third and rarest position could be derived by applying (17)
only. By legislating thus, however, transformational rules seem to be used
to derive less frequent expressions from more frequent ones, and the sig-
nificance of such a procedure is not immediately clear.
In accordance with Canedo's approach the initial position of the Verb
could be partly accounted for by observing that at least in part it charac-
terizes sentences that are derived transformationally from underlying
Phrase-markers containing special transformational markers (e.g., question
111orphemes, etc.). But sentences immediately derived from a base Phrase-
marker (e.g., those with asti or those which, speaking semantically, continue
a story or a description) may also have the Verb in initial position. The
situation with regard to middle position ofthe Verb is similar. On the whole
we shall see that the views of Western Sanskritists would have to be adapted,
if an overall distinction between underlying and derived sentences is adopted.
With regard to the structure of nominal sentences Canedo disagreed with
Delbrueck and Speyer: "Voranstellung des Subjekts [ist] vie! hiiufiger als
Voranstellung des Priidikatnomen" (36). As the most common order for the
(verbal) Active sentence he regards the following disjunction: either 'Sub-
jekt-Objekt-Priidikat' or 'Pradikat-Subjekt-Objekt' (39). Accordingly, two
of the three alternatives mentioned earlier (27) are considered as common
('habituell'). For the Passive he lists: either 'Instrumental-Priidikat-Subjekt'
or 'Instrumental-Subjekt-Pradikat'.
Out of the many views on more specific matters a few more may be
mentioned here. While 'prepositions' and 'preverbs' sometimes precede and
sometimes follow their Noun or Verb (for the prepositions a rule is given,
but it is 'nicht absolut' and 'versagt oft' (44)), thetmesis of the Vedic preverb
receives special treatment. The following example is quoted frequently:
apa tamal.z piipmdnaiJ1 hate 'he strikes [hate] the darkness, the evil away
[apa]' (Taittiriyasm!lhitii 2.1.10). Though this kind of tmesis could also
occur in English (as in the translation just given), Canedo considered it
for Sanskrit as "in der Natur der Sprache bedingt; sie ist eine natlirliche,
56
WESTERN SANSKRITISTS ON WORD ORDER IN SANSKRIT
nicht eine stilistische Eigentiimlichkeit derselben" (88). Thls would prevent
the adoption of a rule such as (15) (above, page 12), unless:
(where Prev denotes 'Preverb' and Bv denotes 'Verbal Base') is added and
a transformational rule is subsequently applied to effectuate the required
rearrangement.
With regard to tbe combimition of a Preposition with a Noun-Phrase
consisting of Noun + Attribute, Canedo mentioned three possibilities:
Attr-Prap-Subst; Attr-Subst-Prap; Subst-Attr-Prap. Only the middle case
would follow directly from the rules of the base (i.e., from (31) on page
51); for the other two additional transformational rules for rearranging
the elements would be required. As regards the order of Noun-Phrases,
Canedo considered the following four as the most common arrangements:
Instr-Lok-Akk-Verb; Lok-Akk-Verb; Instr-Akk-Verb; Dat-Akk-Verb. Note
that they are not inconsistent with each other, and could be formulated as:
{
(Instr)- (Lok)}
- Akk- Verb.
(Dat)
As to the relationship between Genitive and Noun, "eine der verwickeltsten
Erscheinungen der altindischen Syntax" (60), either of the two possible
combinations is found. Whether, finally, an Adjective precedes or follows
the Noun it qualifies, may depend on the meaning of the Adjecti;e (64).
In other words, Adjectives would have to be grouped in two not necessarily
disjunct classes.
Canedo's monograph shows, among other things, that attempts at too
wide generalizations which take only surface structure into account and
which are primarily based upon impressions of frequency, are bound to
result in an enun1eration of various alternatives and numerous exceptions. 3
Within the same framework a much more careful attitude is adopted in
Gonda (1951; 1952), where the problem area is delineated more precisely
while at the same time the material upon whlch the investigation is based is
more varied and extensive. Gouda's work is moreover distinguished by his
consideration of a much greater variety of sentence types. Gonda (1952)
takes exception to Delbrueck's view accepted by Speyer, Thommen and
3 Compare Dover's judgment about studies of Greek word order: "More often, an objec-
tive general statement of the facts appears, to the seeker after rules, inconclusive in the
extreme; it amounts to saying <xyz and xzy occur, but, on the other hand, yxz, yzx, zxy,
and zyx also occur"' (Dover 1960, 1). Greenberg's universals of word order are nearly
always almost-univt:trsals; theY are moreover misleading in that no distinction is made
between sentences with different deep structures. It seems very difficult, lastly, to fit in
Sanskrit (Greenberg 1963, 61-3).
57
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
Canedo, that word order should be studied exclusively from prose material:
"Nous ne possedons dans cette prose gauche, raide et monotone de l'an-
cienne litterature th6ologique et sacerdotale qu'une 'refraction litteraire'
speciale de la langue vivante" (6). The poetic tradition in Sanskrit, on the
other hand, is not "par lit, artificielle" (7). His conclusions are therefore
basJp upon an investigation into various kinds of texts, prose as well as
poetry, Vedic as well as classical Sanskrit, and including the epic, the
drama, and other literary work.
Some of the main results are the following. The Verb occurs most fre-
quently in final position. In metrical klivya and in the however,
it occurs in the middle. Initial position of the Verb is due to a variety of
specific reasons, such as that the sentence is in the Imperative, the Future, or
that it expresses a question, an exclamation, or the intention of the speaker;
that the verb is asti or a verb of knowing, often in the first Person; that the
sentence is a subordinate clause, expresses the of the previous
sentence, of, more general, the continuation of a narrative; or that the Verb
is emphasized. Apparently, most of these cases are derived sentences where
transformations are anyhow required. Middle position of the Verb, as for
example in the texts quoted above, is similarly explained by special con-
siderations, e.g., that other elements are put at the end of the sentence in
order to facilitate transition to the next sentence, or else that the Verb
follows an absolute case or a case which expresses time, place, etc. (cf.
'English 'here lies your book').
Despite these regularities, to which there are almost always exceptions
(cf. above, page 57, note 3), the author arrives at the following clearly
formulated interpretation of the mo.terial: "En ce qui concerne l'ordre des
e!ements de l'enonce, l'indo-europeen comportait une grande liberte, c.-a-d.
quecet ordre n'avait presqu' aucune valeur grammaticale [my italics]. Aucun
mot n'avait dans la phrase indo-europeenne "ne place definie et constante.
11 va sans dire que ceci n'exclut pas !'existence d'ordre habitue! dans certains
tours et certains types de phrases: comme il resulte des recherches qui vont
etre mentionnees, on trouve souvent dans des phrases analogues de plusieurs
langues indo-europeennes anciennes un ordre des mots a peu pres constant
et identique" (Gonda 1952, 71).
In order to further illustrate and support some of these generalizations,
the author finally provides some numerical data regarding the number of
occurrences of the Verb in initial, middle and final position, for selected
passages taken from a variety of texts. In this connection various sentence
types are again distinguished.
Gouda's investigations show that word order, even if it performs no
grammatical function (whatever this means exactly), is relatively fixed in a
58
WESTERN SANSKRITISTS ON WORD ORDER IN SANSKRIT
number of cases. He provides, however, no overall generalization which
accounts for these facts or which leads to a theory about a fundamental or
basic word order. It is not immediately clear, on the other hand, whether a
similar analysis when confined to the types of sentence that can be said to
belong to the base, would not reveal a more regular pattern. For this purpose
a fresh analysis might be required, which would, however, when pursued
along similar lines, at the most yield certain statistical regularities.
Summarizing, we may conclude that Western Sanskritists started to describe
a traditional, habitual or merely preponderant fixed word order in Sanskrit.
When these investigations became more precise, more exceptions were met
with and various habitual arrangements of words for various kinds of
sentence were postulated. It was also increasingly realized that word order
has no grammatical significance or value. But most of the investigations by
Western Sanskritists were based upon a specifically selected corpus of
Sanskrit texts.
59
\
CHAPTER IV
SUMMARY AND CONCLUSIONS
On superficial examination the views on word order in Sanskrit propounded
by Indian theorists and by Western Sanskritists appear to be greatly at
variance. Almost all Indian theorists did, either implicitly or explicitly,
regard word order as free. For no independent word is a specific position
in the sentence prescribed. Sentences which differ in the arrangement of
their words only, are considered as equivalent and synonymous. In so far as
a difference of meaning or different shades of meaning can be found at all,
this is not considered relevant." Most Western Sanskritists, on the other
hand, reject the idea that word order in Sanskrit is free. According to them
there are certain preferential, traditional, habitual or, at least, most frequent
arrangements. Other arrangements, which also occur, tend to be treated as
special cases or as exceptions to an overall regularity. Though more detailed
investigations have shown that there are many different arrangements of
words to account for different types of sentences, there is again a preferential
order within each type. These descdptions, which generally offer approxi-
mations to regularities, if not rules, for the order of words, seem to be
supported by frequency counts.
On closer inspection the gap between Indian and Western approaches
appears to be narrow. This is not to say that the divergence is merely a matter
of language and presentation. The two approaches, although to some
extent complementary, are basically different. They agree howe,e; in that,
on the whole, word order has no grammatical significance.l For the Indian
1
See especially Gonda 1952, 71, quoted above, page 58. For Latin cf. Marouzeau, liT,
1949, 195: "La notion de ce qu'on appelle l'ordre grammatical n'est pas moins trompeuse
[i.e., than 'logical order'}: dans une langue a construction libre, la grammaire ne com-
mandel'ordre que dans des cas exceptiounels, qu'il suffit d'enregistrer,et qui ne comportent
pas d'explication de principe .... " For Greek cf. Dover 1960, 3; "If two utterances are
syntacticaily identical, but differ in order, this does not prove that the determinants of
their different orders are unknowable; it proves only that syntactical identity does not
suffice to determine identity of order; and our task becomes the exploration of all the
respects in which the two utterances are dissimilar, in the hope of finding there the vital
difference which determined their .difference of order" (note that if this task pertains to
utterances, it seems to pertain in particular to performance). Dover's study leads to the
following conclusion: "It is clear that these statistics are very far indeed from establishing
for 'Classical Greek' simpliciter anything worth calling a syntactical rule of word order.
Extended to a much greater variety of authors and texts, they would no doubt give us an
60
SUMMARY AND CONCLUSIONS
grammarians the purport of this doctrine is quite specific: grammatical
relations between words in the sentence, i.e., kiiraka relationships and
similar grammatical relationships are expressed by inflexion and the like.
The order of words of the sentence, on the other hand, has no such signifi-
cance; it is entirely superficial. The relationship between the two approaches
may be further analysed by reminding ourselves of Pataiijali's reaction to
certain opponents who attached significance to word order in the Szltra:
'restrictions on usage' (prayog"aniyama), said Patafijali, 'are not here clung
to' (above, page 28). Similarly, the Mimiil)lsakas agreed with their oppo-
nents, the followers of Kautsa, at least in one respect: even if word order is
fixed in certain utterances this does not imply that word order is
fixed in sentences in the. same manner in which the order of elements .is
fixed, for example, in nominal compounds and inside other words (above,
page 25).
Adapting these criticisms to the interpretation of the Western approach,
we may state that the latter deals with usage and utterances, rather than with
the grammaticality of sentences. This also explains the large numbers of
exceptions, as well as mutual contradictions. Of course, both kinds of study
are important, although the study of usage and utterances tends to be fruitful
only to the extent that the study of grammaticality is presupposed. For
this reason it may be assumed that the Western Sanskritists' investigations
would have led to more definite results, had they adopted the perspective of
a theory of language which distinguishes between underlying or base sen-
tences, and derived sentences. Dover, in his study of Greek word order, is
,clearly aware of the difference between grammatical and frequent: "It is
easy, but wrong, to equate 'statistically normal' with 'natural' and 'statisti- _
cally abnormal' with 'distorted', 'inverted', etc. If, for example, we were
investigating a language in which the order SP was invariable in statements
and the order PS invariable in questions, it would be misleading to formulate
the rule in terms such as: 'SP is normal, but this normal order is in
questions'. State1nents are more numerous than questions, bUt that is not a
fact of a kind with which we are concerned; we should content ourselves
with the pair of discoveries that (a) statement determines the order SP, (b)
interrogation determines the order PS" (Dover 1960, 5).
The difference between the Western and the Indian approach to the study
of Sanskrit word order may be characterized by saying that the former was
mainly interested in the use of language or in performance, while the latter
was mainly interested in the underlying regularities of language or in
interesting picture of the vagaries of individual preference - and thereby. suggest with
increasing force that all patterns of order which are describable in syntactical terms are
secondary phenomena" (ibid., 31).
61
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
competence. The Indian grammarians were accordingly interested in rules,
the Western Sanskritists in probabilities of occurrence and frequencies.
In some cases where Western and Indian scholars arrived at similar results,
this may be explained by observing that the Indian scholars also were
este!;l in the utterances of the Vedic corpus and not in the sentences of the
SanSJcrit (or, as the case may be, Vedic) language. In this connection it
may be noted that the Indian theorists (including men like Apte) were
krit speakers, whereas Western Sanskritists have generally studied Sanskrit
from the corpus of Sanskrit texts, in the same manner in which scholars of
Latin have studied Latin. 2
It is not quite necessary to study Sanskrit as a dead language, confining
oneself to a corpus of texts (even though this corpus is immensely large).
The present use of Sanskrit as a spoken language should be compared less to
the modern, than to the medieval use of Latin. The most widespread view
of Western Sanskritists in this respect is formulated as follows by
nagel-Renou: "Ceux-Ja memes qui etaient familiers avec le Sanskrit parlaient
concurremment une autre langue, plus populaire; le sanskrit avait a peu
pres la position du Iatin au moyen age, de l'Mbreu chez les Juifs. Ajourd'hui
meme, exclusif du sanskrit est tout au plus une gageure; cependant il
serait faux de Iui denier le caractere d'une !angue parlee" (Wackernagel-
Renou 1957, 21-2 and 86-7 notes 310, 312).
Some passages of the Report of the Government of India Sanskrit Com-
mission (1956-1957) are more explicit arid suggestive: "Just as many of the
replies to the Questionnaire received by the Commission were in Sanskrit,
.quite a number of interviews also took place in Sanskrit. It was not the
Pandits alone who gave their evidence in Sanskrit ... " (P: 9); "Even at the
present day Sanskrit is very living, because a large number of people use
Sanskrit in their conversation, when they come from different parts of the
country, and composition in Sanskrit, in both prose and verse, goes on
almost unabated. It has been possible to write a history of recent Sanskrit
literature ... " (p. 88).
Whatever the precise significance of these statements, a theory of usage
accounting for the arrangement of words in utterances has to be based upon
a theory about the grammatical relations between words in sentences. It
is becoming increasingly clear that "investigations of performance will
proceed only so far as understanding of underlying competence permits"
(Chomsky 1965, 10).
2
Cf. Marouzeau 1953, 114 on Latin: "Ce qui nous manque pour assurer notrejugement,
c'est le sens intime de la Iangue tel que le possedaient les sujets parlants; nous y devons
suppleer par une interpretation laborieuse fondee sur des nbgles parfois sujettes a revision
ou discussion."
62
SUMMARY AND CONCLUSIONS
Theoretical considerations apart, there are specific cases where the dis-
tinction between competence and performance becomes explicit. An 'oc-
casional' utterance which represents an occurrence with a low acceptability
or probability value, is not the same as an ungrammatical sentence which
violates a rule (cf. Katz & Fodor 1964, 27). Moreover, though both grammati-
cality and acceptability admit of degrees, their scales do not coincide
(Chomsky 1965, 11). This may be seen from the comparison of simple
examples of sentences from the base in English and Sanskrit. Whereas
apasyad riimo govindam (cf. above, page 12) is perfectly grammatical and
unambiguous, meaning 'Rama saw Govinda', but has perhaps a lower
probability value than, e.g., riimo govindam apaSyat, matters are quite
different with respect to the string Riima Govinda saw, which is quite un-
grammatical in English and in addition meaningless or at least ambiguous
(for it might mean 'Rlima saw Govinda,, 'Govinda saw Rama', 'he saw
Rama Govinda', 'he saw Rama and Govinda', etc.), even though it may
perhaps be met with under exceptional circumstances of English usage.
The difference between apaSyad riimo govindam and saw Riima Govinda is
inexplicable if the theory is adhered to that word order in Sanskrit may be
described by a set of grammatical rules just as can be done in English, even
if it be assumed that the rules for Sanskrit would be more complicated than
those for English.
The difference in this respect between Sanskrit and English may also be
illustrated by considering the interpretation of sentences derived transfor-
mationally from a generalised Phrase-marker. Sentences such as those
quoted by Pataiijali (anar/vtiham udahtiri yii ... : above, page 32) are perhaps
uncommon, but they do not lack grammaticality any more than the similarly
uncommon grammatical English sentence "Anyone who feels that if so
many more students whom we haven't actually admitted are sitting in on the
course than ones we have that the room had to be changed, then probably
auditors will have to be excluded, is likely to agree that the curriculum needs
revision" (Chomsky &Miller 1963, 286). On the other hand both Sanskrit
govindasya rdmo 'paSyan and English saw Govinda Rdma, as -well as sentences
transformationally derived from their underlying base Phrase-markers, are
(though for different reasons) ungrammatical.
While nobody would object to saying that apasyad rtimo govindam and
riimo govindam apasyat are synonymous, it may not be difficult to find
examples of two Sanskrit sentences which only differ in the arrangement of
the words but where a difference in meaning is nevertheless felt. It may also
be felt that two such sentences would not fit into larger contexts equally
well. Gonda often deals with cases where this state of affairs obtains, while
it is referred to explicitly by Marouzeau at the beginning of his study on
63
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
word order in Latin: ".Oeux ordres ossibles ne sent pas synonymes"
(Marouzeau 1922, I). hile such differences may well be explicable within
the context of a theory of performance, it must be noted that they could not
be explained by any theory of competence which utilizes transformations to
e f f e ~ t u t e re-ordering, but adheres to the principle that transformations do
not \tffect meaning. For according to such a theory different arrangements
are generated by applying transformational rules, and these result in
expressions synonymous with the underlying sentences. (Sometimes different
orders may, of course, have to be derived from different deep structures.)
Comparing English and Sanskrit we therefore find ourselves reverting to a
not unfamiliar traditional doctrine, i.e., that Sanskrit expresses by means of
inflexion what English expresses by means of word order as well as (e.g.,
in the case of Pronouns) inflexion (cf., e.g., Bloomfield 1933, 197; Hirt 1925,
66: " ... es ist bekannt wie flexionslose Sprachen durch eine im Wesentlichen
feste Wortstellung einen Teil dessen ausdriicken, was sonst durch die Flexion
erzielt wird"; etc.). This then may be said to be the significance of the state-
ment that word order in Sanskrit is 'free'. The problem arises how to con-
struct a mechanism which first derives expressions for certain grammatical
relations and subsequently provides rules for mapping these expressions into
either ordered sequences of wOrds or strings, or, equivalently, unordered sets
of words which are however determined by inflexion. It may be clear from
our earlier results that such expressions for gram1natical relations may be
derived by means of a system of kiiraka relationships as developed by Pii.J;ini
or, equivalently, by means of Chomsky's definition of 'grammatical relation',
iiiustrated with the help of trees which need not, however, be trimmed. It
may be re-emphasized here that Chomsky was aware of the fact that the
definition of'grammatical relation' does not presuppose order: "the systems
of grammatical relations defined in the two cases [i.e., set-systems and
concatenation-systems] are identical" (Chomsky 1965, 125). This also pro-
vides the background for his remark that "the same grammatical function
may be defined by several different rewriting rules of the base" (Chomsky
1965, 72). We could also express tl1is by saying that wild trees may be re-
garded as equivalence classes of trimmed trees. The relation defining these
equivalence classes is the relation between any two trimmed trees in which
corresponding nodes dominate tl1e same set.
We shaH now give a sketch of how such a system could be set up, making
use of wild trees and company-sensitive rules (above, page 15). Wild trees
are trees such that the points dominated by a node constitute a se.t. Such
trees symbolize the rules of the base, which are now constructed so as to
introduce sets. These sets are stratified, i.e., the hierarchy of subsets containing
elements dominated by a single node, is preserved. Otherwise it would be
64
SUMMARY AND CONCLUSIONS
impossible to identify the constituents and phrases required to determine
conditions for the application of transformational rules; it wonld also be
impossible to derive grammatical relations and give a semantic interpretation.
It should be emphasized that wild trees perform all the functions required
from the trimmed trees familiar from generative grammar, except the capaci-
ty to generate order: they provide the constituents and phrases needed for
the application of transformational rules; they enable us to define grammati-
cal relations; and they may, accordingly, provide us with a basis for the
semantic interpretation of the sentences they picture (note that the projection
rules in Katz's semantic theory correspond with grammatical relations: Katz
1966, 165, 167). They do not, of course, yield the word order which is re-
quired, e.g .. in English. This will be effectuated by special order-introducing
rules, which are analogous to the company-sensitive rules required for the
introduction of concord or government. Since concord is in fact better
described by transformational rules than by context-sensitive rules, transfor-
mational rules operating on sets may be required for the introduction of
order as well (as in (52), below, page 77); but this is inessential in the
present context.
As a fragment of a simplified base we shall now adopt the following rules
(of. above, page 14):
S --+ {NP, VP}
VP --+ {NP, V, Num}
NP --+ {N, Num, Cas, (S)}
(35)
(36)
(37)
Num--+ { ('1' for Sg, '2' for Dual, '3' for PI) (38)
Nom
Ace
Instr
Cas --+ Dat
Ab!
Genit
Lac J
(39)3
(Note that 'Det', 'Person', and mauy other markers and specifications may
be required for a more adequate account).
3
Note that the braces in (38) and (39) express alternatives, whereas in (35)-(37) they
denote the boundaries of sets. In these cases however there is no danger of ambiguity.
65
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
These rules may be applied, for example, as symbolized in the following
wild tree:
(x)
This base can be made a sound point of departure for either an English or
a Sanskrit grammar. Government or concord will be introduced for Sanskrit
by means of company-sensitive rules. When formulating such rules we shall
make use of a triple arrow (=7). Note that company-sensitive rules may be
simpler than the corresponding context-sensitive rules, because not all
elements of a set need be referred to, whereas all adjoining elements of a
string must be referred to.
That the Main Verb is governed by the Subject of S is expressed by the
following triplet:
{{N, i), {V, Nurn}}=7{{N, i),{V, i)) for i = 1, 2, 3. (40)
That the Subject of S is in the Nominative is expressed by:
{{N, Cas}, VP}=7{{N, Nom}, VP}. (41)
That the Direct Object of S is in the Accusative is expressed by:
{{N, Cas}, V}=7{{N, Ace), V}. (42)
More special cases could be similarly accounted for, e.g., the rule that the
Verb is in the Singular if the Subject is a Neuter Noun even when it is in the
Plural. For this purpose, (37) would have to be replaced by:
NP-> {N, Num, Cas, Gend, (S)} (37')
and we would also need:
{
Masc }
Gend-> Fern .
Neutr
The required company-sensitive rule, which presupposes ( 40) and follows it,
is the following:
{{N, 3, Neutr), {V, 3}}=7{{N, 3, Neutr}, {V, 1}}.
66
SUMMARY AND CONCLUSIONS
For English, proceeding from rules of the same base consisting of (35)-
(39), we require company-sensitive rules which introduce ordered strings in-
stead of sets. For example, after (41) we need:
{{N, Nom}, (41*)
and after ( 42):
{{N,Acc}, V"N"Acc. (42*)
Alternatively, if we are interested in English grammar only, and not in
comparative or universal grammar, this may be abbreviated and we may
write instead of(41)-(41*) and (42)-(42*):
{{N, Cas},
{{N, Cas}, V} V"N"Acc.
(41**)
(42**)
In this manner order may be introduced. At the same time, either (41 **)and
(42**), or (41)-(41*) and (42)-(42*), as compared with (41) and (42)express
the generalization that the same relations 'Subject-of' and 'Object-of' are
expressed by case terminations in Sanskrit, and by order as well as case
terminations (e.g., in the case of Pronouns) in English, respectively.
A very simple rule which ought to be expressed is that modifiers in general,
and Adjectives, Participles, etc., in particular, are governed by the Noun they
qualify. In the examples from Sanskrit met with earlier, such sentences were
transformationally derived from underlying sentences, which are nominal
sentences (this is of course not always the case). Such sentences would result
from a base - still simplified but set up in a slightly more sophisticated
fashion (e.g., by introducing 'Pred-Phrase'; see Chomsky 1965, 107)- with
the help of a transformational rule which deletes a Copula or a similar
verbal expression. This may be assumed to result in the expression:
{NP., {NP
2
}}. To express.the required concord we now adopt the triplets:
{{N
1
,i},
{{N
2
,i}}} for i=1,2, 3 (43)
and
{{N
1
, Masc}, {{N
2
,
Masc}, {{N
2
, Masc}}} (44)
(Similarly for Fern and Neuter.)
When these sentences are embedded in Noun-Phrases with the help of (37),
this stratification will have to be preserved under the transformations which
introduce Adjective, etc. Elsewhere it may be necessary to introduce order,
e.g., to distinguish between Adjectivization and Apposition, as follows:
{{N
1
, Masc}, {{N
2
,
{{N
1
, Masc}, {{N
2
,
1
"Masc"N
2
"Masc.
With the help of such procedures we are in a position to state that Adjec-
67
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
lives are governed by the Noun they qualify, but also to distinguish subse-
quently between Adjectives which follow the Noun they qualify, and
Adjectives which precede it (as is required in French and in many other
languages). The general statement that Adjectives are governed by the Noun
they qualify, however, has a much wider scope and is also valid, for example,
for >,sentences where a qualifier occurs which is not specifiable as either
Adjective or Apposition (as, e.g., in Japanese).
The distinction between verbal and nominal sentences is a very important
one in languages such as Sanskrit. In India the status of the nominal sentence,
which is well developed as well as common in scientific Sanskrit (see, e.g.,
Staal 1965c), gave rise to a controversy between the grammarians and the
Naiyayikas (logicians). Whereas the grammarians insisted that the verbal
sentence is basic and that nominal sentences are derived from them by
deleting the Verb (i.e., that they have zero-occurrence of the Copula), the
Naiyayikas considered this a very artificial device and regarded the nominal
sentence as a primary category. This controversy may have resulted from the
exclusive emphasis laid on Noun and Verb by logicians and grammarians,
respectively (Matilal 1966, 381 and 388 sq.).
We have seen (above, page 52) that Delbrueck and Speyer, followed by
other Western Sanskritists, introduced the rule that in the nominal sentence
in Sanskrit the Predicate precedes the Subject. The discussion about the
apparent counter-example embodied in the first rule of Piil)ini's grammar
(above, page 27), points in the same direction. It need not at present be
decided whether this constitutes a grammatical rule relating to conipetence,
or whether it is an observation regarding performance. It may be noted,
however, that in this case word order as a grammatical feature would not be
surprising; for ambiguity might otherwise result, just as in Germau 'Die
Mutter sieht die Tochter'. (The construction of copula sentences in English
may be similarly ambiguous, cf. 'Miss Castlewood was the prettiest girl
at the ball', Jespersen 1948, 153.) In the present context it is only necessary
to show that if the Subject foilows the Predicate because of a grammatical
rule, this may be expressed in terms of the system sketched above. That
order may be introduced is clear from (41*) and (42*) or (41'''*) and (42**);
it is equally obvious that we may introduce order at any stage, wherever it
should be required.
Such a procedure does not mean, however, that whenever order is required
someiVhere, it should be introduced right at the beginning. Yet it is this
assumption which is apparently made in current work on generative grammar.
The assumption underlies the importance attached by Hall to such Russian
sentences as 'Mat' vide/a doe" (1964, 407), and by Chomsky to such German
sentences as 'Die Mutter sieht die Tochter' (1965, 126). These sentences
68
SUMMARY AND CONCLUSIONS
show that word order is not always irrelevant; but they do not show that word
order is never irrelevant, i.e., always relevant. Whatever the precise sitnation
in Russian and German, we now know the position in Sanskrit fairly well.
The derivation of a specific word order for NP"VP sentences would be
irrelevant syntactically; it is, accordingly, counter-intuitive. In order to
avoid ambiguity the Sanskrit grammarians adopted a convention for certain
sentences of the meta-language (above, page 28). Though this does not carry
over automatically to the object-language (cf. na yathti Joke tathti vyakara(re
'it is not so that what applies to ordinary speech applies to grammar itself'),
Pataiijali admits that usage in tllis respect is more widespread, when he
says; neha prayoganiyama tirabhyate 'restrictions on usage are not here
clung to' (above, page 28). But this usage, which is also observed by some
Western Sanskritists, remains exceptional. Similarly, sentences such as
'Die Mutter sieht die Tochter' are traditionally regarded as exceptions to the
rule that word order is syntactically irrelevant (cf. above, page 60 note 1:
"la grammaire ne commande l'ordre que dans des cas exceptionnels"). The
most natural syntactic theory to account for this, will treat them as such.
They can be easily accounted for by introducing order, when required, into
a system that is otherwise in this respect unordered. Before discussing further
implications of such an approach, let us see how precisely the required order
could be introduced.
Adopting in the base (still simplified, cf. Chomsky 1965,' 107) the rule:
VP-> {{Copula; NP})J
{V, NP}
we require along with (40)-(42) (abo.ve, page 66):
(45)
4
{{N, Nom}, {Copula, {N, Cas}}}=H{N, Nom}, {Copula; {N, Nom}}}
(46)
If the order Predicate-Subject were a grammatical feature, it could for
Sanskrit be introduced by the same rule which introduces nominal sentences,
i.e., by:
{{N
1
, Nom}, {Copula, {N
2
,
(47)
It would seem conceivable that an analysis of word order in Sanskrit,
similar to the investigations carried out by Western Sanskritists, but con-
fined to sentences immediately derived from a base Phrase-marker, would
unexpectedly reveal that also for verbal sentences the order at least of some
words is grammatically fixed. This is extremely unlikely in view of all we
4
The braces denote alternatives and set boundaries; cf. above, page 65 note 3.
69
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
have said before, and also because a fixed order such as this would possess
no syntactic or semantic function (as it has in nominal sentences) and would,
e.g., introduce a distinction between riimo govindam apayat and apaSyad
govindal[l riimab, which is never made. But even if this eventuality should
come true, it would still be just another exception. Moreover, we would still
a general rule which expresses that sentences, whether nominal or
vefbal, consist of a Noun-Phrase and a Predicate-Phrase, i.e., an unordered
rule:
S-+ {NP, Pred-Phrase}.
This could only be replaced by a rule introducing an ordered string, if the
arrangement in all cases to be just the same. But this is even more
unlikely, since in nominal sentences the Subject was determined-as habitually
following the Predicate, which is unlikely in most other cases.
Apart from Sanskrit and quite in general, all such considerations tend to
show that systems incorporating wild trees and company-sensitive rules
are much more likely to adequately express linguistic universals than rules
which presuppose or introduce order. This would be so; even if word order in
Sanskrit were utterly fixed - if we except the case of its being fixed in the
same way as word order is fixed in English or in any other language, a
result which no amount of abstraction could bring about. Let us on the one
hand assume, in accordance with this hypothesis, that the order which perhaps
occurs most frequently in Sanskrit, i.e., rtimo govindam apayat (cf. above,
page 12), were grammatically determined, i.e., by the rules:
S -+ NP"VP (14)
VP-+ NP''V. (15)
For English, on the other hand, no legerdemain could convince us that we
ought to dispense with:
S -+ NP"VP
VP-+ V"NP.
(I)
(2)
Though (1) happens to coincide with (14), (2) does not coincide with (15),
and hence the linguistic universal behind (2) and (15) can only be expressed
by:
VP-+{NP, V}.
This may be subsequently specified in different ways for different language
types, as for English by:
{NP, V"NP
but for Sanskrit, in accordance with this hypothesis, by:
{NP,
70
I
I
I
!
I
i
SUMMARY AND CONCLUSIONS
If, however, a language existed where we have a grammatical rule:
S-+ VP"NP (1")
(as the nominal sentences in Sanskrit perhaps suggest), we must similarly
replace (1) or (14) by:
S-+ {NP, VP}. (1')
Accordingly we arrive at a system of universal grammar which is basically
unordered and can only be described in terms of sets and wild trees. Order
may be intmduced into this unordered base as early as a specific language
type permits. This conclusion does not depend on the conclusion arrived
at earlier, i.e., that word order in Sanskrit is 'free'. For even if free word
order in this sense would not obtain anywhere (as Delbrueck 1878, 76 and
Chomsky 1965, 126 were inclined to think), the divergent arrangements
of words in different types of language would require the base of a system of
universal grammar to be unordered.
Of course, this generalization can also be made on the basis of Chomsky
1965; but only in a redundant and rather unnatural way. One might simi-
larly adopt a medical meta-language, constructed in such a way that it can
naturally express that men hav.e a heart on the left side or on the right,
and observe subsequently, that this implies that men have a heart. Medical
theorists rightly adopt a language which enables us to express straight
away that man has a heart.
Some scholars have found it difficult to understand that a system of gener-
ative grammar is not intended to serve as a model of the speaker or hearer
(e.g., Reichling 1961; 14, 16; Uhlenbeck 1963, 9-10; but see Kraak 1966,
especially 44-47). This error, which is the outcome of the systematic refusal
to deal with deep structure which characterizes taxonomic linguistics, perhaps
arises less easily when the base of a generative grammar is unordered in the
way indicated. For it is obvious that no one could want to claim that people's
brains contain some unordered structure representing the set {NP, VP},
which is then somehow converted by neural or other mechanisms into the
ordered string NP"VP (for some) or into VP"NP (for others). The assump-
tion that people are aware of the underlying set when or before they utter
an ordered string, is even more fantastic. Such views would be similar to
claiming that people's bodies contain a subtle heart, which is located no-
where, but which is converted into a gross manifestation on the left or on the
right side. It is clear, on the other hand, that systems of generative grammar
are systems of linguistic description, a particular instance being thaf (1')
is postulated to account for both (1) and (1"), which in turn are to account
for a variety of structures, including surface phenomena. Such postulates
71
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
have, again, nothing to do with a system being 'logical' or influenced by
mathematical logic, as the same- and other- critics have complained.
It may be noted that the adoption here suggested of unordered sets to
replace ordered strings, is similar to that adopted in Chomsky 1965, where
the ,ordered branches of the semantic trees of Katz and Postal (1964) have
bee!\ replaced by complex symbols denoting sets of syntactic features. - Of
course, it is not suggested that this state of affairs (or that referred to in the
previous paragraph) need in any way be regarded as an argument in support
of the set-systems here advocated.
It is far beyond the scope of the present monograph to extend this in-
vestigation effectively to universal grammar, which would presuppose the
availability of generative grammars. for a variety of natural languages. We
shall have some more general comments to make on universal grammar,
but shall first consider some other examples in order to further illustrate and
corroborate the above results. We have already seen that in both Latin and
Greek the situation is very different from that in English, in that word order
has on the whole no syntactical significance. For Latin, Marouzeau stated
that grammar determines the order of words only in exceptional cases (see
above, page 60 note 1). For Greek, J22Y!'! suggested that "all patterns of
order which are describable in syntactical terms are secondary phenomena"
(ibid.). The Sanskrit system may in principle be described by ( 40)-( 42)
(above, page 66) and (45)-(47) (above, page 69). With regard to nominal
sentences, the. situation in Arabic and Hebrew (and perhaps in Semitic in
general) is the exact opposite. Here too Western scholars appear to have
combined investigations into competence with studies of performance.
Lacking the competence to unravel this performance we may be excused
for merely quoting Fleisch (1956, 122; cf. BrocRelmann 1956, 118-121):
" .. . L'arabe classique ne demande rien ala place des mots pour caracteriser Ieur
fonction dans la phrase; par les voyelles finales de la declinaison (et de la
conjugaison), il avait en effet le moyen de determiner, d'une m n i e r ~ in-
herente au vocable, sa fonction dans la phrase. Et cependant l'ordre des
mots n'est pas libre. En dehors des combinaisons rigides- comme determine
suivi du determinant dans l'annexion grammaticale, adjectif epithete apres
le nom- il existe un ordre a respecter: verbe + sujet + complement direct
+ complement circonstanciel [i.e., in verbal sentences]; sujet + attribut +
complement circonstanciel, en phrase noininale."
Singling out for further illustration the arrangement of Subject and
Predicate, we would have to express this statement, provided it describes
grammatical reguladty, by (45), and in addition, instead of (40), by:
{ {N, i}, {V, Num}} =? V'Y'N"i for i = 1, 2, 3 (48)
72
I '
.I
I
I
I
'
'
t
SUMMARY AND CONCLUSIONS
and instead of ( 47), by:
{{N
1
, Nom}, {Copula, {N
2
, Cas}}}=o>N
1
'"'Nom'"'N
2
'"'Nom. (49)
Comparing the nominal structures to the right of the symbol '=o>' in ( 47)
and (49) for Sanskrit and Arabic, respectively, it is obvious that no expression
of universal grammar which could conceivably be inserted to the left of
':=;/, would be able to account. for these two sentence structures, unless it
were formulated in terms of sets. Moreover, the general production of a
purely nominal sentence is given again in terms of sets, i.e., by:
{{N
1
, Nom}, {Copula, {N
2
, Cas}}}=?{{N, Nom}, {N
2
, Nom}}. (50)
Tlus must be a transformational rule, since there is deletion of an ele-
ment (cf. below, page 77).
With regard to Arabic two further observations may be made; both
concern exceptions which help to establish the general rules.
First, the habitual word order became syntactically significant in middle
Arabic when the terminations of classical Arabic had been lost (as in the
parallel developments in Indo-European and, perhaps, in Archaic Chinese).
As a reaction, authors of the middle Arabic period, e.g., Mutanabbl (916-65
A.D.) employed a completely free word order when writing classical Arabic.
This was syntactically correct, though in some respects unusual. Even the
most severe critics of Mutanabbl and of his language, however, never ob-
jected to this (Fiick 1955, 148-50).
The second case concerns concord between Verb and Subject in NP"VP
sentences. In general, concord depends exclusively on grammatical relations
and must therefore be expressed by rules like (40) (above, page 66). As
always in Indo-European, it depends on grammatical relations defined in
terms of the surface structure ('he is easy to please'), not on the deep structure
of underlying base Phrase-markers, from which it is transformationally
derived (which would account for 'hun is easy to please'). Thus concord
depends for example on the grammatical subject, not on the logical subject
(cf. above, p<tge 66; Chomsky 1965, 221-2 note 35). But in Arabic not the
surface structure, but the arrangement of words in the finally derived string
may in exceptional cases influence concord, esp,cially when in tllls string the
Predicate precedes the Subject. In that case, says keckendorf (1898, 69) "ist
vom Subjekt vorerst nur eine vage Vorstellung vorhanden" (note that this is
formulated with reference not to competence, but to _performance). This
explains, e.g., cases where a verb in the plural precedes a subject in the inner
plural (which otherwise determines the singular): ja!,ulfina 'abna'u a/-ba'iri
'spoke [Plural] the sons [Inner Plural] of the camel'; or cases where a verb
73
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
in the singular precedes a subject in the dual: tantiza'ani ar-raguliini 'fought
[Singular] with me the two men [Dual]'.
In English the selection of Pronouns generally depends in a similar manner
on the arrangement of words in the finally derived string. Lastly, some
more examples, partly from English, will be considered. It seems fairly
certAin that (3) or (3'), which have to account for all the recursive properties
oflanguage (except conjunction and the like, but cf. Rosenbaum 1967), will
be utilized for the introduction of Adjectives, Appositions; Participles, other
sorts of Nominalizations, and all kinds of phrases. A certain number of
these will require a rearrangement of words. For example, if (3) is adopted
for Adjectivization by means of embedding, either the second occurrence of
the resulting two identical Nouns will have to be erased, or the order of N
and S must be reversed; for restrictive clauses, however, the order may be
preserved. If (3') on the other hand is adopted for introducing restrictive
cla\1ses, the order ofN and S must be it may be preserved, however,
in the case of Adjectivization.
Many similar examples could be given for English, whereas in the case
of many subordinate clauses it is not even clear which order we want to
describe, e.g., 'though he loved smoking, he gave it up' as against: 'he gave
up smoking, though he loved it'.5 Of course, even in cases where there is no
need to change the order, transformational rules will still be wanted for
generating the required structures.
It is quite conceivable that in the of a moro detailed description
of such structures, arguments for the adoption of either (3) or (3') will
come forward. But even if these arguments would predominantly point into
one direction (which seems unlikely), it is obvious that only an unordered
rule would allow a universal procedure to be followed. For, while (3) or
(3') require the transformational rules to preserve order in some cases but to
alter it in others, transformationai rules for all cases determine the required
order, provided they are applied to the result of the application of an un-
ordered rule such as:
NP-> {(Det), N, (S)}.
5
From an accoUnt given by Mr.R.P.G. de Rijk of recent work on transformations
carried out at M.I.T., I learned that it is always possible to replace in the finally derived
string the second occurrence of two NP's with the same reference by a Pronoun (e.g.,
'John noticed that he was becoming deaf'). It is sometimes possible in English (but never, it
seems, in Turkish or in one of the languages of Ghana) to replace the first occurrence of such
an NP by a Pronoun, provided a certain condition is fulfilled (e.g., 'That he was becoming
deaf annoyed John'). This condition depends on grammatical relations and may therefore
be formulated in terms of sets. Thus, there are two kinds of Pronoun which are quite
different in origin (note that also in English the first kind, but not the second, may be
replaced by 'the aforementioned').
74
SUMMARY AND CONCLUSIONS
Moreover, only this rule would express the generalization that for English
(not to mention other languages) recursiveness is introduced into sentences
by embedding other sentences into Noun-Phrases.
It may well be asked how transformations work in a syntactic theory
based upon sets. It seems obvious that permutations are only definable in
terms of strings. This may seem relevant in connection, e.g., with the analysis
of questions. But just as it may seem preferable to describe questions not
as permutation transformations from affirmative sentences, but both af-
firmative sentences and questions as transformations from underlying
abstract structures (just as positive and negative sentences are derived from
such structures, rather than negative sentences from positive ones: Kraak
1966, 158), it seems preferable to describe such.abstract structures in terms
of sets rather than strings, and apply transformational rules to these sets.
In the case of deletion and addition transformations it is even more
likely that this is in many cases the most promising solution. It is obvious that
in Sanskrit the optional deletion of tena 'by him' from such sentences as (24)
(above, page 40):
tena kumbhdb kriyante 'pots are made by him',
should be defined in such a way that it also applies to the other five grammati-
cal permutations, i.e., kumbluis tena kriyante, tena kriyante .kumbhiib, etc.
This can only be done by means of one rule, if the description is given in
terms of sets.
Passivization itself, in Sanskrit or Latin, cannot be described in tenus of
strings either, if it is to be described in a natural and not counter-intuitive
way. In Sanskrit, rdmo govindam apaSyat 'Rama saw Govinda', and each of
its five permutations, is grammatical, and no particular one is optimal.
But each of the permutations of: adrsyata rtimeva govinda/:l 'Govinda was
seen by Rama', is equally grammatical and is the Passive of each of the
former group of six. To express this in terms of Passive transformations
formulated in terms of strings would require quite a large number of rules.
One rule will again account for this, provided Passivization is described in
terms of sets.
In general this would require that the Analyzability condition for the
applicability of transformational rules be in such cases expressed in terms of
sets. An example from the literature may make this clear. For the sake of
exposition and simplicity, agreement has in this monograph been generally
described by means of context-sensitive or company-sensitive rules. It may be
preferable, however, to describe it with the help of transformations. If this
is done as in Katz and Posta1 (1964, 9-11), the situation depicted in their
Diagram 2.4:
75
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
Sentence
Noun Phrase Verb Phrase
I
Noun
A
Verb Adverb
~
A Nounstem
Gender Number
1 2 3 4
but not that depicted in their Diagram 2.5:
Sentence
Noun !Phrase
~ u
Verb Phrase
A
Adverb Verb
. A NounStem
Gender Number
2' 3 4
falls under the domain of the structure of the transformation:
Gender+ Number, Noun Stem, Verb, X=1, 2,1 + 3, 4.
1 2 3 4
This is clearly stated on page 10 (lines 3-4). But is it desirable? It seems
obvious, on the contrary, that this structure index carries redundant infor-
mation in the same way as some context-sensitive rules: for we want to deal
with the situation depicted in Diagram 2.5 as well, since the position of the
Adverb and the order of words in general has nothing to do with the agree-
ment of the verb with the subject (cf. above, page 9, where a similar re-
76
SUMMARY AND CONCLUSIONS
mark was made concerning the description of grammatical relations). The
redundant information precisely prevents the additional application of the
transformational rule which is reqnired, and thus imposes too strict limits
on the applicability of a transformational rule. We need instead a kind of
transformational rules defined on Analyzability conditions similar to the
company-sensitive rules introduced earlier.
It is no doubt possible to replace the Analyzability conditions for the
applicability of transformations formulated in terms of strings by similar
conditions formulated in terms of sets, analogous to the replacement of
context-sensitive by company-sensitive rules. In order to do this we may
have to mark the elements of sets in a similar way as was earlier employed
when stratification of sets was indicated by braces.
The company-;ensitive rule for verbal agreement given above ((40) on
page 60) can in fact be written as a transformational rule with a structure
index, i.e.:
{{N, i), {V, Num)}.,.{{l, 2}, {3, 1)}.
1 2 3 4
(51)
On the other hand such order-introducing rules as (47) (above, page 69)
may be written as transformational rules of the form:
{{N, Nom}, {Copula, {N, Cas}}}=4"2"1"2. (52)
1 2 3 4 5
By such simple changes the whole Of transformational theory is preserved;
but attention should be paid to the precise point where in grammars for
individaallanguages, rules which generate strings are introduced.
We have now seen that even in English there are cases where it is more
appropriate to formulate transformational rules in terms of sets than in
terms of strings; the change, moreover, is largely terminological. Gram-
matical relations, as we have seen repeatedly, are always inde;>endent of
order, and so is of course the system of constituents itself. The same holds
for the semantic interpretation, since each of the projection rules of semantic
theory coincides with a particular grammatical relation (see above, page 65).
The arrangement of words in a sentence is therefore only semantically
significant (as it is, e.g., in English) in as far as this arrangement expresses
grammatical structure (as it does in English).
The resulting system is relatively simple but it is not as simple as the system
of Chomsky 1965. The reason is that we were forced to introduce unordered
rules to start with. From here one may proceed by introducing order along
Chomskian lines, but in different directions for different structures within the
same language, or for different types of language. Any system of universal
77
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
grammar appears to require such an unordered base, and so do systems of
generative grammar for languages with 'free word order'.
If we wish to describe in more precise terms in what way such a system
of universal grammar is more complicated than the system of Chomsky
1965, we shall see that its greater complexity is primarily due to the fact that
Chd,msky's system constitutes in fact a special case, appropriate for languages
with a fixed word order. This explains at the same time that certain rules of
Chomsky's system appear to carry redundant information from the point
of view of universal grammar. Since even in English adverbials move rather
freely, it is not certain that there are languages with absolutely fixed word
order. Let us in general distinguish theoretically between (1) ~ ~ l a Q g ~ s
with free word order; (2) B-languages, with pa_rtiai!YJree = partially fixed
word order; and (3) C-Iangl!llges, with fixed word order. In theoretical
terms it may be postulated that universal grammar contains a substructure
of the base of each grammar of an A-, B- or C-language. It might in fact
contain the following rules:
S --> {NP, VP}
VP ->{V, NP}
NP--> {N, (S)}
(a)
(b)
(c)
and furthermore, apart from other kinds of universals, rules which introduce
other constituents such as Prep-Phrase, Sentence Adverbial, Prep-Phrase
. Adverbiai, and perhaps Det Such. a base embodies the universal deep
structure and is mapped into a variety of particular surface structures. But
this mapping is done by different rules for different languages or language
types. These added rules (which are largely transformational) constitute for
each language a particular 'supplement', which is not universal and does not
contribute to the semantic interpretation. Of course, these supplements
may again overlap or contain other supplements, and thereby circumscribe
certain language types. For A-languages, their rules derive further sets; for
C-languages, they immediately generate strings; for B-languages, a combina-
tion of both.
Speaking theoretically, (a)-(c) belongs to every language. But if (a)-( c),
thus formulated in terms of sets, is regarded as a fixed subpart of each
particular grammar, such a grammar, especially if it describes a C-language,
may turn out to be more complicated than it need be by itself. There is,
however, no objection to replacing the sets of (a)-(c) immediately by the
required strings, if one is engaged in the practical job of writing a grammar for
a C-language, provided one has no general theoretical or comparative
purpose as well. That would enable us to arrive at precisely the system of
Chomsky 1965.
78
. I
J
I
. l
J
SUMMARY AND CONCLUSIONS
This might be expressed in different terms as follows. The adaptation of
particular grammars of C-languages by replacing the sets of the universal
base immediately by strings, signifies that the entire edifice of one universal
grammar along with all the particular supplements which make up particular
grammars, is the simplest possible. But this does not imply that a particular
grammar made up of the universal grammar along with one particular
supplement, need also be the simplest possible. This is not surprising from
a methodological point of view .
It seems most likely that most languages are in fact B-languages, though
some may be almost similar to A-, and others almost similar to C-languages.
While English may be almost a C-language, Sanskrit appears to be almost
an A-language. The entire edifice of univerGal grammar as embedded in a
particular grammar for Sanskrit acquires a structure which is basically
Pal)inian: it is a system which first introduces grammatical relations analo-
gous to the kiiraka relations (at the same time providing for the possibility
of transformations and for semantic interpretation); subsequently introduces
order on the basis of these relations; and finally introduces such elements
as certain Pronouns, as well as sandhi rules, which operate on ordered
strings.
The system here envisaged may also be contrasted with earlier systems
and characterized as follows.' So far, generative grammar has been conceived
of as a device consisting of a syntactic component with two interpretations,
one a phonological component and the other a semantic component. The
ordered rules of the base introduce strings of formatives, constituents and
dummy symbols, thus generating generalized Phrase:markers represented by
trimmed trees. A)1lorrg the rules of the base there is at least one recursive
rule, in which the initial sentence symbol S is allowed to reappear on the
right, so that the rules may reapply to this new occurrence of S. The trans-
formational subcomponent of the syntactic component maps the deep
structure generated by the b&se into surface stn:ctures. Whilst most dummy
symbols are signals indicating that rules of the transformational subcompo-
nent have to be applied (i.e., transformational markers), the symbolS, which
6
In order to avoid all misunderstanding it may be emphasized that the system here d ~
vacated does not intend to explain free word order by replacing 'trimmed' trees by stratified
'wild trees'; this attempt would fail, e.g., for Sanskrit, for we have seen (above, pages 34,
54) that words may move beyond the boundaries of substrings dominated by particular
constituents, or (which means the same) beyond the subsets of stratified sets. This failure
itself is one of the arguments which favours the assumption that word order cannot be
described in terms of trees, whether trimmed or wild. Therefore, the conclusion is drawn
that word order does not belong to the deep structure, but must be relegated to the surface
structure. But once this conclusion is reached, there is no more need to cling to trimmed
trees, and trimmed trees may be everywhere replaced by wild trees without any" loss.
7
The remaining paragraphs owe their existence to a stimulating discussion with Dr. Kraak.
79
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
may be considered both a dummy and a constituent symbol, signals that
a Phrase-marker must be embedded. The semantic interpretation, lastly, de-
pends on the system of constituents and grammatical relations defined
by the base and is derived by inserting the lexical items of the lexicon in
positions in the base and applying projection rules which correspond
to trammatical relations.
In the theory here envisaged the base of the syntactic component consists
of a kernel of ordered rules (including at least one recursive rule) introducing
sets of formatives, constituents and dummy symbols (transformational
markers). This kernel is the same for all natural languages and hence a
linguistic universal which is presupposed by particular grammars. Since the
kernel generates stratified sets, which may be represented by wild trees, it im-
mediately provides the system of constituents and grammatical relations
required for the semantic interpretation. The kernel of the syntactic com-
ponent, therefore, coincides with the non-lexical subpart of the semantic
component, i.e., with a system of projection rules. Whereas the rules of the
base in the system of Chomsky 1965 were "to a large extent universal"
(ibid., 141), the rules of the kernel are precisely the universal rules which
determine the semantic interpretation. In addition to the kernel, the syn-
tactic component contains other rules which vary from language to lan-
guage or from language type to language type. For some languages these
rules derive further sets from the sets generated by the kernel; for other
languages they immediately generate ordered strings .. It is clear that in
neither case these additional rules contribute to the semantic interpretation.
It is probable that these rules are transformational rules which operate on ..
sets. Ultimately ordered strings will be introduced everywhere, and surface
structures will be derived from these with the help of other transformational
rules. The picture which emerges is not that of a syntactic component with
a semantic interpretation (deep structure) and a phonological interpretation
(surface structure), but that of a kernel consisting of stratified sets which
embodies the universal deep structure and is mapped into particular surface
structures.
Essentially such a theory is much more simple and perspicuous than
earlier systems, since it permits the generation (or 'interpretation') of
structures in one way only, viz., from the unordered deep structure expressed
by the kernel to ordered surface structures. This is intuitively more satis-
factory and constitutes a natural development from the basic ideas that
prompted earlier formulations of theories in generative grammar.
80
APPENDIX
SOME SEMANTIC RELATIONS
As an illustration of a specific application of the theory outlined above,
some semantic relations recently treated (Staal 1967a), partly within the
framework of Chomsky's Aspects, will be reconsidered. My solution of
. 1967a, if appropriate at all, will be seen to be so only for English or other
languages with similar, at any rate fixed, word order. The special devices
introduced there depend on word order and surface structure, and are
therefore inherently superficial. The particular arrangement of words which
happens to be grammatical in English was utilized only to the extent that it
provided information about the underlying grammatical relations, which
are themselves independent of word order. In other words, these special
devices introduced redundant information in the same way as context-
sensitive rules (see above, page 11). The required generalization is accom-
plished by replacing strings by sets and trimmed trees by wild trees; this can
be carried out in a relatively simple manner. It may be assumed that this
replacement is a step towards the generalization from English syntax to
universal grammar.
In order to clearly show the point at issue I shall first briefly sketch the
solution proposed in 1967a; subsequently show that this solution is not
applicable to similar cases from Sanskrit; and finally introduce modifications
which allow the theory to apply to both English and Sanskrit (as well as
some other languages). It may be assumed that the resulting description
belongs to universal grammar.
The problem under discussion is how to analyse the relation of paraphrase
obtaining between such sentence pairs as:
John precedes Mary
Mary follows John
(in at least one sense of the verbs), or:
Peter buys books from John
John sells books to Peter.
(53)
(54)
(55)
(56)
The following solution was proposed. Special indicatory symbols were
introduced into the complex symbols which occur in lexical items. For
81
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
example, the lexical item for precede was written in the lexicon as:
(precede, [+V, +--NP<fol/ow) ... ]).1
In general, if a verb is given in the lexicon in the form:
)
(V, [+V, +-"(Prep;)''NP' <V'"(Prep.)) ... ]),
(57)
(58)
this was defined so as to imply that the following sentences are paraphrases:
NP"Aux"V" ... "(Prep;)"NP'
NP'"Aux"V'" ... "(Prep.)"NP.
(59)
(60)
The actual result of this device is that a noun-phrase which, in a terminal
string, immediately precedes the verb, and another noun-phrase which
follows the verb but is together with the verb dominated by VP, change
places. But what ought to be effectuated is that the subject should change
position with another specified noun-phrase (e.g., the noun-phrase of the
object), which is dominated by VP, but which need not follow the verb. This
is the requirement as formulated in terms of grammatical relations and not
in terms of accidental surface structures. In English, the subject happens to
precede the verb immediately and the other noun-phrase happens to follow
the verb. But in general this is not so (i!l some cases there are alterr.atives
also in English, as we shall see.)
This becomes obvious as soon as we consider an analogous example
from Sanskrit (actually almost any other language would provide similar
indications). In Sanskrit we have to account for the relation of paraphrase
which obtains between the following sentences (a specific word order,
different from English, has been oelected):
indrab somam upajivati 'Indra lives on Soma' (61)
soma indrmfl dhtirayati 'Soma supports Indra'. (62)
The underlyine Phrase-markers may be represented by the trees:
NP V
soma upajivati
s
~
NP VP
soma
/ ~
NP
indra
V
dhfirayati
(xi)
1
The braces'{','}' adopted in 1967a are replaced by angles '<',')', since braces are used
here to refer to sets.
82
APPENDIX: SOME SEMANTIC RELATIONS
where the occurrences of V are simplified in that indications of person,
number, tense, etc. have been omitted.
We shall for the sake of simplicity denote verbs in the lexicon by means
of the specific representative quoted here, i.e., by the third person of the
present tense indicative. Let us assume, accordingly, that the lexicon contains
the item:
(upajivati, [+V, + __ NP <dhiirayati) ... ]). (63)
Applying definition (58), the relation of paraphrase between the following
sentences may be accounted for:
indra upajivati somam
somo dhdrayatindram.
(64)
(65)
However, each of the six permutations of (61) or (64) is a paraphrase of
each of the six permutations of (62) or (65). To account for this it is not
enough to interpret the trees of (xi) as wild and not as trimmed trees,
unless (63) is also modified. Moreover, if upajivati is a transitive verb, this
cannot in general be expressed by specifying that the verb is followed by a
NP (though dominated along with this NP by VP), i.e., by a context-
restriction:
+_NP, (66)
since this restriction is confined to surface structures similar to those of
English and cioes uot apply to a number of other language' for which it is
undoubtedly appropriate to say that they possess transitive verbs. We must,
therefore, introduce a more general 'company-restriction', which specifies
that the verb is accompanied by an object or combined with an object
(which may precede or follow it). Precisely this results from the restriction:
+{_,NP}. (67)
In general, the context-sensitive subcategorlsation rules involving expressions
like (66), introduced by Chomsky (1965, 90 sq.), are not valid for Sanskrit
or universal grammar. They should be replaced by company-sensitive
subcategorisation rules involving expressions like (67). While this is an
important difference from the conceptual point of view, the formal change
is easily effected by replacing strings by sets and '"' b!f ', '.
The lexical item for upajivati may now be written as follows:
(upajivati, [+V,+{-, {NP<dhiirayati, + {_, NP}) }}]). (68)
In general, if a verb is given in the lexicon in the form:
(V, [+V,+{-, NP' <V',+{_, NP})}l) (69)
83
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
this is defined so as to imply that the following structures underlie sentences
that are paraphrases:
{NP, {V, NP'}}
{NP', {V', NP }}.
(70)
(71)
App.,Jying this to (68), and further utilizing (37) (simplified), (41) and (42)
a b o ~ e , page 66) we can derive the following structures, which underlie
sentences that are paraphrases:
for example:
{{N, Nom}, {upajivati, {N', Ace}}}
{{N', Nom}, {dhiirayati, {N, Ace}}}
{indrab, { upajivati, somam}}
{ soma/1, { dharayati, in dram}}
and hence (61)-(62), (64)-(65), etc.
(72)
(73)
(74)
(75)
The device introduced by the definition of (69) not only accounts for the
required paraphrases in Sanskrit, but in English as well. The lexicon will
now be required to contain the item:
(precede, [+V,+{_, NP(follow, + {-, NP})}l). (76)
This implies that the following structures underlie sentences .that are para-
phrases:
{NP, {precede, NP'}}
{NP', {follow, NP }}.
(77)
(78)
If we now utilize (37) (simplified), (41**) and (42**) (above, page 67) we
can account for the paraphrase relation between .the following structures:
N "Nom"precede" N' "Ace
N' "Nom" follow" N" Ace
(79)
(80)
(the verbal terminations may be added and specified with the help of (40)-
above, page 66- and other rules, which have not been taken into account in
order to simplify the present discussion). This accounts for 'he precedes her'
and 'she follows him' being paraphrases, and also for the paraphrase
relation between (53) and (54).
This treatment of paraphrase relations is more complicated than that
proposed in 1967a. But this earlier simplicity was achieved at the cost of
excluding general features of language which are manifested in English in
a special way. It is, of course, fully legitimate to exploit the idiosyncratic
properties of English for an English grammar not written for comparative
purposes or for the sake of exhibiting general features of language. Still, the
advantage of the present treatment (and also of, e.g., (67) over (66)) appears
not merely from the point of view of universal grammar (as these particular
84
APPENDIX: SOME SEMANTIC RELATIONS
English structures are now derived from structures in universal grammar),
but also from the point of view of English grammar, as these English struc-
tures are now derived from what they really depend on, i.e., not surface
structure, but underlying grammatical relations. In view of this it may be
assumed that independent syntactic motivations for such descriptions can be
found also in English. 2
We must also account for the paraphrase relation between (55) and (56)
(above, page 81), which involves more complicated devices. The underlying
Phrase-markers may be taken to be, respectively:
s
NP
VP
/
N
V NP
Peter
buys books
and:
books
A
Prep NP
I
from !
to
N
John
(xii)
Peter
(xiii)
2
Transitive verbs are sometimes preceded by the.ir object, with or without inversion of
the subject and main verb of the sentence: 'Many a rabbit had he snared without the game--
keeper noticing it'; 'Dates I could never remember'; 'Invalids we have no use for' (quoted
by Zandvoort 1961, 282).
85
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
The solution proposed earlier and expressed by means of (58) can also be
applied here, provided the lexicon contains:
(buy, [+V, + __ NP"jiom"NP (sell"to)]). (81)
Ac<;ording to the definition this-implies that (55) and (56) are paraphrases.
Here again, the noun-phrase whlch immediately precedes the verb and the
noun-phrase which immediately follows from, change places. But what ought
to be described is that the subject changes places with the noun-phrase
immediately dominated by Prep-Phrase, itself immediately dominated by
Verb-Phrase. This is the requirement as formulated in terms of grammatical
relations and not in terms of accidental surface structure.
we can now provide the required subcategorization for the verb buy.
In the system of Aspects, this subcategorization would be given in terms of
the context-restriction:
+ __ NP"from"NP. (82)
But this does not apply in a number of other languages where the same
graramatical relation underlies the same phenomenon. We need instead the
company-restriction:
+ {_, NP, {Prep, NP)}. (83)
The English lexicon will accordingly be required to contain the following
lexical item:
(buy, [+V,+{_, NP, {fi'om, NP(se/1, + {_, NP, {to, NP)})}}l)
(84)
which will in a more natural and justifiable manner account for the paraphrase
relation between (55) and (56).
In Sanskrit we have to explain the relation of paraphrase between the
following sentenc"s (a specific word order, different from English, has
again been selected):
devadatto rtimtiya rukmal]1 dadtiti
'Devadatta gives a golden ornament to Rama' (85)
rtimo devadatttid rukmam tidatte
'Rama receives a golden ornament from Devadatta'. (86)
The underlying Phrase-markers may be represented by the following wild
trees, respectively:
86
APPENDIX: SOME SEMANTIC RELATIONS
s
NP VP
/
N
/ ~
K, V
devadattaQ
/ ~
dadiiti
NP Dat NP Ace
I I
N N
riim5ya rukmam
(xiv)
and:
s
~ ~
N
;p ~ ~
K
5
Kl V
/ ~ / ~
iidatte riimal)
T Ab! T Ace
N N
devodottllt rukm!Jm
(xv)
where the occurrence of V is again simplified by omitting person, number,
etc. In these trees the symbols K; have been introduced to remind ourselves
of Pal)ini's terminology, where karaka relations (cf. above, pages 36-45)
are said to obtain between the verb and the different noun-phrases of the
verb-phrase. The underlying karaka relation for riimiiya is sarrzpradiina,
which is manifested by the Dative or 4th case (in Pill).ini's order); the
underlying kiiraka relation for rukmam is karman, which is manifested by
the Accusative or 2nd case; the underlying kiiraka relation for devadattiit is
apadiina, which is manifested by the Ablative or 5th case. In the above
terminology, the kiiraka relations obtain between V and the NP immediately
dominated by K;. The K; themselves express in general what was expressed
by Prep-Phrase in the examples from English. But the constituent K; may
87
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
be realized as {Prep, NP}, {Cas, NP}, as a NP in a specified place in the
terminal string, or in other ways. In this system, the Object-of relation,
casu quo the kmman kiiraka relation, casu quo the Accusative case, has
been included. Moreover, the Subject-of relation, casu quo the Nominative
case, could have been included as well.
The lexical item for dadtiti may now be written as follows:
(dadati, [+V,+{_, K
2
, K
4
(tidatte, + {_, K
2
, K
5
} )}]). (87)
We shall postulate that the symbol occurring inside as well as outside
the angles (i.e. K
2
, in this case) will not be involved in the change of position
which is being described. Provided we use (instead of (37): above, page 65)
certain rules of the type:.K,-->{NP, Cas;}, and in addition rules to specify
the cases and phonological rules, we can account for the paraphrase relation
between (85) and (86) by adopting the following general convention.
If a verb is given in the lexicon in the form:
(V, [+V,+{_, Kj (V',+{_; Kk}) }l) (88)
this is defined with reference to cases (though it will be differently defined,
e.g., with reference to prepositions) so as to imply that thefollowingstructures
are paraphrases:
{{NP, Nom}, {V, {NP', Casj}}} (89)
{{NP', Nom}, {V', {NP, Cask}}}. (90)
When applying this to (87) we can derive from (89):
{{NP, Nom}, {daddti, K
2
, {NP', Cas
4
}}} (91)
{{NP, Nom}, {dadtiti, {NP, Ace}, {NP', Dat}}}; (92)
and from (90):
{{NP', Nom}, {Matte, K,, {NP, Cas
5
}}} (93)
{{NP', Nom}, {Matte, {NP, Ace}, {NP, Ab!}}}. (94)
This accounts for the paraphrase relation between (85) and (86), or any
other permutation of the words of these two sentences, respectively.
For English we start again from (88), which is now defined with reference
to Prep-Phrases so as to imply that the following structures underlie sentences
that are paraphrases:
NP" Aux"V" ... "(Prep;)"NP' (59)
NP'"Aux"V'" ... "(Prepk)"NP. (60)
This accounts for the paraphrase relation between (55) and (56), provided
we start from the lexical item:
(buy, [+V,+{_, NP, {from, NP (sell,+{_, NP, {to, NP}})}} l).
(95)
88
APPENDIX: SOME SEMANTIC RELATIONS
Since (60) yields only the sequence: V"NP""to"NP', we may also stipulate
that another rule be framed which generates the equivalent order:
V"NP'"NP". This will account for 'John sells Peter books'.
In the formalisms introduced here, Cases, like Prepositions, are combined
with Noun-phrases, and not with Nouns as was done before (see (37),
above, page 65). This seems fully justified in the present context, though
further investigation is required before a particular solution can be justified
in general.
The replacement of strings by stratified sets, which is the new feature
introduced by these descriptions, may have to be worked out differently
when more material is taken into account. Whatever its final shape, the
introduction of stratified sets serves the purpose of formulating grarmnatical
descriptions, whether syntactic or semantic, directly and exclusively in terms
ofthe grammatical relations which are involved. If these grammatical relations
were formulated in the manner first proposed by Chomsky, i.e., as relations
between, or ordered pairs of, constituents belonging to different levels of a
tree, it would be difficult to introduce these relations directly into the de-
scriptions. Nowhere does Chomsky in fact make direct use of these formal-
izations of grammatical relations. The grammatical relation object-of, for
example, defined as [NP, VP] (Chomsky 1965, 71) cannot be easily ir.tro-
duced into a subcategorization rule since the constituents which occur in
these expressions were introduced by different re-writing rules, i.e., occur on
different levels of the tree:
VP
A
(xvi)
V NP
If, on the other hand, the relation object-of is required for the definition e.g.
of transitive verbs, it is more adequate, as we have seen, to use the expression:
{_,NP}. (96)
This expression exactly characterizes the relation between, on the one hand,
the V which fits into the open place, and, on the other hand, the entire
expression itself in as far as it represents the VP directly dominating it. In
other words, expressions such as [NP, V] may be eliminated altogether,
for stratified sets such as { __ , NP} characterize the corresponding gram-
matical relations directly. This is even more obvious when we consider a
grammatical relation between constituents belonging to levels of a tree still
further removed from each other. The grammatical relation between the
NP realized as Peter, immediately dominated by Prep-Phrase which is
89
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
immediately dominated by VP in (xiii) (above, page 85), could not be
expressed by [NP, VP], since this relation already defines the relation
object-of This relation is however precisely characterized by the stratified
set:
{V, NP, {Prep, -}}. (97)
)
As noted before (page 35), the expressions for stratified sets are derived
from wild trees by using a familiar method, i.e., in the same manner in which
(labelled) bracketing is derived from trimmed trees (the braces which mark
the boundaries of sets could, of course, also be labelled, if necessary):
s
~
NP A
V NP
trimmed tree
(NP"'(V"NP))
bracketed string
s
~
/ ''-
NP ~
V NP
wild tree
{NP, {V, NF}}
stratified set
(xvii)
The illustrations discussed in this appendix further support the view that
some of the syntactic er semantic properties earlier described in terms of
the word order or surface structure of terminal strings, are more adequately,
more naturally and more generally described in terms of the grammatical
. relations on which they depend, i.e., in terms of stratified sets.
I
90
BIBLIOGRAPHY
A. TEXTS AND TRANSLATIONS
A$!lidhytiyi of Pa:oini, ed. and transl. by 0. Boehtlingk, Reprint Hildesheim 1964.
-, Transl. by L. Renou, I-ID, Paris 1948-1954.
KiiSikiivrtti of Jayaditya, ed. by Sobhita MiSra, Banaras 19523.
-, Transl. 1,1: by Y. Ojihara and L. Renou, Paris 1960-1962.
Tantravlirttika of Kumarila Bhatta, ed. (with Mimii1]1siisfitra and
in the A.nandaSrama Sanskrit Series, Poona 1929-34.
-, Trans1. by G.N.Jha (Bibliotheca Indica, 161), Calcutta 1924.
Nirukta ofY8.ska with Durga's commentary, ed. by H. M. and R. G .Bhadkamkar, Bombay
1918-1942.
Brhaddevatii of Saunaka, ed. and transl. by A. A. Macdonell (Harvard Oriental Series),
Cambridge, Mass., 1904.
Mahiibhii$Ya ofPatafijali, ed. by F.kielhorn, 1-III, Bombay 1880--1885.
Miinameyodaya ofN<lnlya}Ja, ed. and transl. by C.Kunhan Raja and S.S.Suryanarayana
Sastri, Madras 1933.
Mimiimsiisrltra of Jaimini: see Tantraviitttika.
Mimiimsiisiitrabhii$ya of Sa bar a: see Tantraviirttika.
-, Transl. by G.N.Jha (Gaekwad's Oriental Series), Baroda 1933-1936.
f$.gvidlziina, transl. by J.Gonda, Utrecht 1951.
J!.kpriitiSiikhya with Uvata's commentary, ed. and transl. by M.D.Shastri,Allahabad 1931.
Viikyapadiya of Bhartrhari, I, ed. and transl. by M. Biardeau, Paris 1964.
SarvadarSanasamgraha of Miidhava, ed. by V.S.Abhyankar, Poona 1951.
-, Transl. by E.B.Cowell and A.E.Gough, Reprint Varanasi 1961.
SiddhiintakaumudiofBhattoji ed. and transl. byS.C..Vasu, Reprint Delhi-Vara-
nasi-Patna 1962.
B. OTHER WORKS
Alien, W. S. (1953), Phonetics in Ancient Inclia, London-New York-Toronto.
Apte, V. S. (196325), The Student's Guide to Sanskrit Composition. A Treatise on Sanskrit
Syntax for the Use of Schools and Colleges, Varanasi.
Bei"ge, C. (1958), Thtiorie des graphes et ses applications, Paris.
Bloch, J. (1906), La phrase nominale en sanskrit, Pc:.ris.
Bloomfield, L. (1933), Language, New York.
Brockelmann, C. (1956), Hebriiische Sylltax, Neukirchen.
Brough, J. (1953), 'Some Indian Theories of Meaning', Transactions of the Philological
Society, 161-76.
Buiskool, H. E. (1939), The Tripiidi, being an abridged English recast of Piirvatriisiddham,
Leiden.
Canedo, J. (1937), Zur Wort- und Satzstellung in der alt- und mitte!indischen Prosa, GOt-
tingen.
Cardona, G. (1965), 'On Translating and Formalizing PiiQinian Rules', Journal of the
Oriental Institute (Biroda), 14, 306-14.
Chomsky, N. (1957), Syntactic Structures
1
's-Gravenhage.
91
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
(1964), Current Issues in Linguistic Theory, The Hague.
(1965), Aspects of the Theory of Syntax, Cambridge, Mass.
(1966a), Cartesian Linguistics. A Chapter in the History of Rationalist Thought, New
York-London.
(1966b), 'Topics in the Theory of Generative Grammar', in Current Trends bz Lin-
guistics m, 1-60.
G. A. Miller (1963), 'Introduction to the Formal Analysis of Natural Languages',
in Handbook of Mathematical Psychology, New York, u, 269-322.
Curry, H. B. (1961), 'Some Logical Aspects of Grammatical Structure', in Structure of
Language and its Mathematical /.spects, Providence, 56-58.
Delbrueck, B. (1878), Die altindische Wortfolge aus dem <;atapathabriihmat;a dargestellt
(Syntaktische Forschungen, Ill), Halle.
- (1888), Altindische Sylllax (Syntaktische Forschungen V).
Dover, K. J. (1960), Greek Word Order, Cambridge.
Edgerton, F. (1928), 'Some Linguistic Notes on the MimiiiJ.sft System', Language, 4, 171-7.
(1929), The Mfmiilisii Nyiiya PrakiiSa or Apadevf, New Haven-London.
Emeneau, M. B. (1955), 'India and Linguistics', Journal of the American Oriental Society,
75, 145-53.
Faddegon, B. (1918), The Vaife#ka System described with the help of the oldest texts,
Amsterdam.
(1936), Studies on Piil;ini's Grammar, Amsterdam.
Fleisch, H. (1956), L'arabe classique. Esquisse d'une structure linguistique, Beyrouth.
Fowler, M. (1965), 'How ordered are PB:!)ini's rules?', Journal of the American Oriental
Society, 85,44-7.
Frauwallner, E. (1960), 'Sprachtheorie und Philosophic im Mahabh8:$yam des Patafijali',
Wiener Zeitschriftfiir die Kunde Slid- und Ostasiens und Archiv fiir indische Philosophie,
4, 92-118.
Flick, J. (1955), <Arabiya. Recherches sl!r l'histoire de la langue et du style arabe. Trad.
C. Denizeau, Paris.
Gonda,.J. (1951), La place de la particu/e negative NA dans la phrase en vieil indien, Leiden.
- (1952), Remarques sur la place du verbe dans la phrase active et moyenne en Jangue
sanscrite, Utrecht.
Government of India Report of the Sanskrit Commission 1956-1957 (1958), Delhi.
Greenberg, J. H. (1963), Universals of Language, Cambridge, Mass.
Hall, B. (1964), Rev.iew of S. K. Saumjan and P.A. Soboleva, Applikativnaja poroida-
juSCaja model' i isCislenie transformacij v russkom jazyke, in 40, 397-410.
Hiriytinna, M. (1932), Outlines of Indian Philosophy, London.
Hirt, H. (1925), Geschichte der deutschen Sprache, Mlinchen.
Jespersen, 0. (1948), The Philosophy of Grammar, London.
Jha, G. N. (1942), PUrva-Mfmiirilsii in its S{)urces, Benares.
Katz, J. J. (1966), The Philosophy of Language, New York-London.
and J. A. Fodor (1964), 'A Reply to Dixon's ''A Trend in Semantics"', Linguistics, 3,
19-29. .
and P. Postal (1964), An Integrated Theory of Linguistic Descriptions, Cambridge, Mass.
Kielhorn, F. (1876), Kiityiiyana and Pataiijali, Osnabrtick (reprint 1965).
Kraak, A. (1966), Negatieve Zinnen. Een Methodologische en GrammaOsche Analyse,
Hilversum.
Kunhan Raja, C. (1952), In Defence of Mfmiilhsii, Adyar, Madras.
Kunjunni Raja, K. (1963), Indian Theories of Meaning, Madras.
Lees, R. B. (1963 2), The Grammar of English Nominalizations, The Hague.
Levin, s: R. (1965), 'Langue and parole in American Linguistics', Foundations of Lan-
guage, 1, 83-94.
Marouzeau, J. (1922-1953), L'ordre des mots dans la phrase latine, I-IV, Paris.
Matilal, B. K. (1960-1961), 'The Doctrine of kiiral)a in Grammar and Logic', Journal of
the Ganganatha Jha Research Institute, 11. 63-9.
92
BIBLIOGRAPHY
(1966), 'Indian Theorists on the Nature of the Sentence (viikya)', Foundations of Lan-
guage, 2, 377-93.
Minard, A. (1936), La subordination dans la prose vedique, Paris.
Oertel, H. (1930), Zur indischen Apologetik, Stuttgart.
Ojihara, Y. and L. Renou (1960---1962): see under A., KiiSikiivrtti.
Ore, 0. (1962), Theory of Graphs, Providence.
Pos, H. J. (1932-1933), 'Ober den Aufbau der grammatischen Interpretation', Bliitter /fir
Deutsche Phi/osophie, 6, 295-314.
Puri, B. N. (1957), India in the Time of Patiinjali, Bombay.
Reckendorf, H. (1898), Die symaktischen Verhiiltnisse des Arabischen, Leiden.
Reichling, A. (1961), 'Principles and Methods of Syntax: Cryptanalytical Formalism',
Lingua, 10, 1-17.
Renou, L. (1940), La Durgha!avrtti de Sarm;adeva. Traite grammatical en sanskrit du XIIe
siecle, I, 1 : Introduction, Paris.
(1947), Les ecoles wJdiques et la formation du Veda, Paris.
see under A.,
(1951), Sarikara: Pro/egomenes au Vediinta. Paris.
(1955), 'Les nlpiitana-sfi.tra de P3.Qini et questions diverses', in Etudes Vediques et
Piil;tneennes I, Paris, 103-30.
(1956), Histoire de la langue sanskrite, Lyon-Paris.
(1957), Terminologie grammaticale dll sanskrit, Paris.
(1960), Le destin du Veda dans /'lnde (Etudes Vediques et VI), Paris.
(1961), 'Grammaire et poetique en Sanskrit"' in Etudes vedlques el Pii{lineennes VTII,
Paris, 105-31.
(1963), 'Sur le genre du sfi.tra dans la litterature sanskrite', Journal Aslatique, 251,
165-216.
(1967), 'PaQini', in Current Trends in Linguistics, V: Linguistics in South Asla, The
Hague-Paris.
and J. Filliozat (1953), L'inde classique II, Paris-Hanoi.
Rocher, R. (1964a),"'Agent" et "Objet" chez PfiQini', Journal of the American Oriental
Society, 84, 44-54.
(1964b), 'The Technical Term hetu in PfiQini's A.JtiidhyO.yi', Vishveshvaranand Indolo-
gical Journal, 2, 31-40.
(1965), 'La formation du futur periphrastique sauskrit selon un exemple de
description linguistique', Annali. Istituto Orientate di Napoli, 6, 15-22.
Rcsenbaum, P. S. (1967), 'Phrase Structure Principles of English Complex Sentence
Formation', Journal of Linguistics, 3, 103-18.
Sarup, L. (1920), The Nigha{l{U and the Nirukta, Introduction, London etc.
Subrahmanya Sastri, P. S. (1960-1962), Lectures on PataJijali's I-VI,
Thiruvaiyaru.
Saumjan, S. K. (1965), 'Outline of the Applicational Generative Model for the Description
of Language', Foundations of Language, 1, 189-222.
Scharfe, H. (1961), Die Logik lm Mahiibhii,Jya, Berlin.
Sk6ld, H. (1926), The Nirukta. Its place in 0/dlndianLiterature-lts Etymologies, Lundetc.
Speyer, J. S. (1896), Vedische und Sanskrit-Symax (Grundriss der Indo-Arischen Philo-
logie und Altertumskunde 1,6), Strassburg.
Staal, J. F. (1961), Nambudiri Veda Recitation, The Hague.
(1962), 'Negation and the Law of Contradiction in Indian Thought', Bulletin of the
School of Oriental and African Studies, 25, 52-71.
(1963a), EuclidesenPiiJJini. Twee Methodische Richtlijnen voor de Fi/osojie, Amsterdam.
(1963b), Review of Scharfe (1961) in Journal of the American Oriental Society, 83,
252-6.
(1965a), 'Context-sensitive Rules in Foundations of Language, 1, 63-72.
(1965b), 'Generative Syntax and Semantics', Foundations of Language, 1, 133-54.
(1965c), 'Reification, Quotation and Nominalization', in Logic and Philosophy. Essays
93
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
in honour of I. M. Bocheiiskl, Amsterdam, 151-87.
(1965d), 'Euclid and P<ioini', Philosophy East and West, 15, 99-116.
(1966a), 'Room at the Top in Sanskrit. Ancient and Modern Descriptions of Nominal
Composition', lndo-lranian Journal, 9, 165-98.
(1966b), Reviews in: The Journal of Symbolic Logic, 31, 245-51.
(1966c), 'Par.dni tested by Fowler's Automaton', Journal of the American Oriental
Society, 86, 206-9.
'Indian Semantics, 1', Journal of the American Oriental Society, 86, 304-11.
(1967a), 'Some Semantic Relations between Sentoids', Foundations of Language,
3, 66-88.
(1967b), 'Sanskrit Philosophy of Language', in Current Trends in Linguistics, V.
Thieme, P. (1931), 'Gramrnatik und Sprache, ein Probiem der altindischen Sprachwissen-
schaft', Zeitschrift fiir Indologie und Iranistik, 8, 23-32.
(1935a), and the Veda, Allahabad.
(1935b), zu varttika 5 zu Pal)ini 1.1.9 und seine einheimischen Erklarer. Ein
Beitrag zur Geschichte und Wiirdigung der indischen grammatischen Scholastik',
Nachrichten von der Gesel/schaft der Wissenschaften zu GOttingen, Berlin, 171-216.
(1956), 'P<ll)ini and the PaiJ.iniyas', Journal of the American Oriental Society, 76,1-23.
Thommen, E. (1903), Die Wortstellung im Nachvedischen, Altindischen und im Mittel-
indischen, GUtersloh.
Uhlenbeck, E. M. (1963), 'An Appraisal of Transformation Theory', Lingua, 12, 1-18.
Varma, S. (1929), Critical Studies in the Phonetic Observations of Indian Grammarians,
London.
Wackernagel, J. and L. Renou (1957), Altindische Grammatik: Introduction Generale,
GOttingen.
Zandvoort, R. W. (1961), A Handbook of English Grammar, Groningen.
94
Abhyankar, V. S. 91
Alien, W.S.l7,91
Apte, V. S. !, 11-2, 21, 62, 91
Aristotle 44
see: PJ)ini
Atharvaveda 17, 24
Bazell, C. E. ix
Benfey, Th. 51
Bergaigne, A. 51
Berge, C. 15, 91
Bhadkamkar, H. M. and R. G. 91
Bhartrhari vii, 5, 91
Bhattoji 19, 42, 91
Biardeau, M. 91
Bloch, J. 55, 91
Bloomfield, L. 64, 91
Boehtlingk, 0. 29, 37, 91
Bonebakker, S. A. ix
Brainerd, B. ix, 15
Brhaddevatii 17, 21, 25, 91
Brockelmann, C. 72, 91
Brough, J. 17, 91
Buiskool, H. E. 3, 91
Canedo,J. 55-57,58,91
Cardona, G. 27, 91
Chomsky, N. passim
Cowell, E. B. 48, 91
Curry, H. B. 13-4, 92
Delbrueck, B. 51-3, 56, 68, 11; 92
Denizeau, C. 92
Dover, K. J. 57, 60-1, 72, 92
Du Marsais 13
Durga 17, 25
Edgerton, F. 20, 92
Emeneau, M. B. 17, 92
Faddegon,B. 3, 37-8,44,92
Filliozat, J. 39, 93
INDEXES
A. NAMES
Fleisch, H. 72, 92
Fodor, J. A. 63, 92
Fowler, M. 29, 40, 92
Frauwallner, E. 44, 48-9, 92
Flick, J. 73, 92
Gonda, J. 25, 55, 57-60, 63, 91, 92
Gough, A. E. 91
Greenberg, J. H. 57, 92
Hall, B. 14, 68, 92
Halle, M. ix
Haradatta 29
5
Hiriyanna, M. 20, 49, 92
Hirt, H. 13, 64, 92
HiZ, H. 13
Jaimini 21, 45, 91
Jayaditya 91
Jespersen, 0. 68, 92
Jha, Q. N. 20, 91, 92
Kaiyata 29
KiiSika (vrtti) 32, 91
Katre, S. M. 24
Katy8yana 19, 30, 32
Katz, J. J. 12, 42, 63, 65, 72, 75-6, 92
Kiutsa 24--5, 45, 47, 49, 61
Kielhorn, F. 5, 19, 27, 32-3, 41, 43--4,
49, 53, 91, 92
Kraak, A. ix, 71, 75, 79, 92
Kumarila Bhatta 21, 47, 91
Kunhan Raja, C. 20, 48, 91, 92
Kunjunni Raja, K. 48, 92
Lees, R. B. 2, 9, 92
Lennep, D. F. W. van ix
Levin, S. R. 46, 92
Macdonell, A. A. 26, 91
95
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
Madhava 91
Mahiibhiirata 55
see: Pataiijali
Miinameyodaya 48, 91
Marouzeau, J. 60, 62-4, 72, 92
Matilal, B. K. 39, 92-3
Mill"f, G. A. 35, 63, 92
MimiN]1sii 19, 21, 24---5, 45-9, 61
Mimdmsii-sUtra 21, 45, 47, 49, 91
Mimii1]1Sii-siitra-bhii.yya 21, 45, 47, 91
Minard, A. 55, 93
Mutanabbi 73
Nage.Sa (Nagoji) Bhatta 19
9!
Niritkta 11, 21, 24-6, 43, 45, 48-9,91, 93
Ojihara, Y. 29, 91, 93
Ore, 0. 15, 93
PadamGJijari 29
Piir:tini 2-3, 17-9, 21, 23, 25-46, 64, 68,-
87, 91-4
Patafijali I, 5, 18-9, 27-8, 30, 34, 37,
41-4, 48-9, 53, 61, 63, 69, 91-3
Pos, H. J. vii, 93
Postal, P. M. 2, 12, 42. 72, 75-6, 92
Pradipa 29
PriitiSiikhya 17, 22, 24
see: Mimiimsii
Reckendorf, H. 73-4, 93
Reichling, A. 71, 93
Renou, L. 1, 5, 16-7, 23-4, 26-7, 29-30,
32, 37-40, 42-3, 45, 49, 62, 91, 93-4
17, 21-3, 25, 48
.fl.gvidhiina 25, 91
Rijk, R. P. G. de 74
R,.kpriitiSiikhya 17, 21-3, 91
Rocher, R. 30, 39, 41, 93
Roser.baum, P. S.-4, 5, 74, 93
Sabara 21, 45-7, 91
Siimaveda 17, 25
Sru\kara 20, 93
Sarup, L. 17, 93
SarvadarSanasamgraha 47, 91
SatapathabrtihmO!Ja 51, 55
Saumjan, S. K. 14, 93
Saunaka 91
SayaQa 25, 48
Scharfe, H. 27, 49, 53, 93
Shastri, M. D. 91
Siddhiintakaumudi 19, 42, 91
Skold, H. 28, 93
Smith, A. 13
Speyer, J. S. 53-7, 68, 93
Staal, J. F. 5, 7, 12-3, 16-7, 26-7, 29,
40, 42-4, 48-9, 53, 81-90, 93-4
Subrahmanya Sastri, P. S. 28, 93
Suryanarayana Sastri, S. S. 48, 91
Taittiriya-sat{lhitti 56
Tantravtirttika see: Kumiirila Bhatta
Thieme, P. 24, 28, 30, 37, 39, 49, 94
Thommen, E. 55, 57, 94
Uhlenbeck, E. M. 71, 94
Uttara-Mimtit[lsii see: Vediinta
Uvata 17, 23, 91
VaiSe$ika 44, 92
Vtikyapadiya see: Bhartrhari
Vardhamana 47
Varrna, S. 17, 94
Vasu, S. C. 91
Vediinta 19-20
58
V os, A. L. ix
Wackernagel, J. 62, 94
Wesly, P. G. E. ix
Whitney, W. D. 2, 3, 16, 37
Yajurveda 17
Yaska 91
Zandvoort, R. W. 85, 94
B. TECHNICAL TERMS: SANSKRIT
adhikaral)a 38-9
adhyaya 18
ananupiirvyasarphita 24
anubandha 27
anuvrtti 19, 40
apadana 38-9, 41, 87
apaSabda 46
96
abhisarnbandha 1-2, 13. 28, 30, 32, 36
avyayibhava 29
48
adeSa 31
anupiirvi 1, 22, 28
iinupiirvya 1, 23, 29, 32
INDEXES
anupUrvyasarphita 24
45-6, 61
upasarjana 43
karaQ.a 38-9
kartar 38-9
karman 36-42, 87-8
kamacara 28
karaka 3, 36-45, 61, 64, 79, 87-8
krt 41-3
kramapiitha 22-3
guoa 27
chandas 18, 26, 29
taddhita 41-2
tin 41-2
dvandva 25
niyata 46-7
pada I, 20
padapiitha 21-4
19, 20
pada 18, 47
prakrti 46
pratyaya 46
pradhiina 43
prayoga 28, 61, 69
bahuvrihi 41
18, 26
yathakamam 28
liighava 19, 40
lopa 32
vakya 1, 20, 45, 47
varttika 19, 25, 32, 41
vikrti 22
vrddhi 27
sarphit<lp,tha 21-4
sarpjii.fi 27
sarpjii.in 27
sa111dhi 11, 17, 20-3, 36, 54, 79
sru:p_nidhi 48
saq1pradana 38-9, 87
sarpbandha 1-2, 13, 20, 30, 36, 43
samii.sa 25, 41-2, 46
samarthya 40
sGtra 4-5, 16-8, 26--32., 38-43, 47
sth3nin 31
sthanivadbhiiva 31-2
svatantra 29
hetu 38-9
C. TECHNICAL TERMS: ENGLISH
analyzabi!ity condition 31-2, 75-7
base 4-6 et passim
company-restriction 83, 86
company-sensitive rules 15, 65- 7, 69-70,
75, 77, 83
competence 61-4, 72
complex symbols 72, 81-4, 86, 88
context-restriction 83, 86
context-sensitive rules 4, 9-11, 15, 65,
75, 83
deep structure 3, 37 et passim
functor 13
genotype 14
grammatical relations 8-9, 14-5, 37-9,
43-4, 64-5, 73, 77, 79-82, 85-6, 89-90
kernel 80
labelled bracketing 34-5, 90
pe:formance 61-4, 72
phenogrammatics 14
phenotype 14
recursive rules 4-6, 33-6, 53--4, 74-5, 79-80
stratified sets 35-6, 66-7; 69, 72-3, 77, 80,
89-90
subcategorization rules 83, 86, 89
surface structure 3, 78-80, 82, 85-6
taxonomy 2, 8, 71
tectogrammatics 13
transformational markers 7, 56, 79
transformational rules 4, 7, 9-12, 40-4, 52,
56-7, 63-5, 67, 74-80
(trimmed) trees 6, 15, 33-5, 51-4, 76, 81-3
wild trees 15, 34-6, 80-3, 86-7
97
WORD ORDER IN SANSKRIT AND UNIVERSAL GRAMMAR
D. LANGUAGES
Arabic 72-4
Archaic Chinese 73
English 7-12, 63-4, 67-8, 70, 74--7, 81-2,
8t-9
French 68
German 13, 68-9
Greek 57,60-1,72
Hebrew 72
98
lndo-European 58, 73
Japanese 68
Latin 13, 60, 62-4, 72, 75
Russian 68-9
Sanskrit passim
Turkish 74
Vedic 17-8, 20-6, 29-30, 45-9, 58, 62