PĀṆINI'S GRAMMAR AND COMPUTER SCIENCE

Author(s): Saroja Bhate and Subhash Kak

Source: Annals of the Bhandarkar Oriental Research Institute , 1991-1992, Vol. 72/73,
No. 1/4, Amrtamahotsava (1917-1992) Volume (1991-1992), pp. 79-94
Published by: Bhandarkar Oriental Research Institute

Stable URL: https://www.jstor.org/stable/41694883

REFERENCES
Linked references are available on JSTOR for this article:
https://www.jstor.org/stable/41694883?seq=1&cid=pdf-
reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.

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

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

is collaborating with JSTOR to digitize, preserve and extend access to Annals of the
Bhandarkar Oriental Research Institute

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 PACINI'S GRAMMAR AND COMPUTER SCIENCE
By

Saroja Bhate & Subhash Как

Introduction

The enterprise of computer science has two fundamental elements.

The first element is to develop techniques that make the elucidation of the
computational structure of nature and the mind easier. The second element
is the creation of new computing algorithms and machines that have powerful
cognitive and computational abilities : this includes development of new techni-
ques of representing and manipulating knowledge, inference and deduction.

The tasks of representing and processing knowledge with a somewhat

different emphasis has parallels in many ancient disciplines. Thus gramma-
rians have long considered questions of relating facts about the physical world
and cognition to linguistic expressions. Likewise logicians have developed
formal structures to relate events and draw inferences from them. This is seen
best in the work of ancient Indian logicians and grammarians. It has been
argued by Ingalls, Staal, Matilal, Briggs, Как and others1 that many contem-
porary developments in formal logic, linguistics, and computer science are a
rediscovery of the work of these ancient masters. But apart from the ques-
tion of a correct history of ideas it raises the following important question of
significance to Sanskritists as well as cognitive and computer scientists : Are
there other rules in ancient Indian logic and grammar that may be of use in
making further advance in cognitive and computer sciences ? A little bit of
history shows why this is a valid question. Nineteenth century Western lingu-

1 D. H. H. Ingalls, Materials for the Study of Navya-Nyãya Logic . Cambridge :

Harvard University Press, 1951.
Frits Staal, Universais , Studies in Indian Logic and Linguistics . Chicago:
University of Chicago Press, 1988.
В. K. Matilal, The Navya-Nyãya Doctrine of Negation. Cambridge ; Harvard
University Press, 1968.
R. Briggs, " Knowledge representation in Sanskrit and artificial intelligence Al
Magazine , vol. 6, 1985, pp. 22-38.
S. С. Как, " The Paninian approach to natural language processing International
Journal of Approximate Reasoning , vol. 1, 1987, pp. 117-130
S. Ç. Как, Thç Riddle of SUnkhya. Delhi ; Allied, 1991 (in press )t

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
80 ABO Rix Amrtamahotsava Volume

ists did not see the significance of the context-sensitive rule

mmar. In fact their fundamental importance was seen only
style structures were first introduced by Western linguists
about thirty years ago. According to the distinguished ling
" We can now assert, with the power of hindsight, that India
fifth century B. C. knew and understood more than Wester
nineteenth century A. D. Can one not extend this conclusio
it is probable that Indian linguists are still ahead of their W
and may continue to be so in the next century ? Quite poss
say is that it is difficult to detect something that we have n
vered ourselves."

Computationally, grammars of natural language are as p

computing machine. But since the setting of a grammar is s
the typical purpose of a computer, this fact is often obscu
structure of a grammar can be easily adapted so as to perfor
cessing. In this paper we discuss formal aspects of certain r
grammar, Astãdhyãyí ( A ), which is traditionally studied to
dhãtupãtha , which is a list of verbal roots arranged into s
ganapatha , which is a list of various classes of morphs, on
dhãtupãtha . It is now becoming clear that A does not merel
lysis of words ( sabdãnmãsana ) but in fact provides a structure f
of sentences. Due to its algebraic nature and its comprehensi
ture has been described as a machine generating words a
Sanskrit.3 Composed in the succinct sütra style, A consists
sutras that capture the fundamentals of Sanskrit language
phonology, morphology and syntax. As in any formal system
consists of definitions, theorems ( linguistic facts ), and meta
regarding rules ). The rules are of different kinds : some a
context-sensitive transformations, others operate sequentiall
Generally these rules are expressed in three groups : ( i ) rules of
or meta-rules samjnã and paribhãsã rules, (ii ) rules prescribi
two kinds of basic dhatu and prãtipadika roots, and (iii) rul
mation for the stems and the suffixes - the morphophonem
that a computer program has exactly the same general feat
sensitive rules, recursion, and sequential rule application. It

2 F. Staal, 47.
3 S D. Joshi, " Sentence structure according to Panini in Glim
Vyãkarana, edited by G. V. Devasthali. Bombay : Popular Prakash
Sf Ç. Как, op. cit .

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 Bhate & Как : Panini' s Grammar and Computer Science SI

ing, therefore, that these sutras have been compared to a computer program
that generates Sanskrit sentences. Panini's grammar is algebraic where a
finite set of rules generates an infinite number of words and sentences.

It is generally agreed that the Pãninian system is based on a principle

of economy, an Occam's razor. This makes the structure to be of special
interest to cognitive scientists. Furthermore, development of logic has been
seen as emerging from the background of grammatical categories both in India
and in Greece. Considering the preeminent position of the Pãninian system
in the Indian intellectual tradition, its significance for students of logic and
history of science becomes clear.

It is also important to place Panini's grammar in the context of a

continuing development of mathematics and science in India. Seidenberg has
shown4 that the rise of the earliest mithematics should be seen in the Vedic
literature. Furthermore, Как has established1 that the Brahmi script of
Pänini's time is to be derived from the Indus script of the third millennium
В. C. This means that Panini himself was heir to a very long and rich tradi-
tion of learning.

Grammatical categories serve to express knowledge about the world.

Panini's system of knowledge representation is based in the kãraka theory.
The kãrakas are deep structure relations that mediate mappings from seman-
tic relations ( such as agent, goal, location) to phonological representations
( in terms of case-endings that may express voices ) via surface structures ( in
terms of morphological categories such as nominal cases, prepositions, and
verbal voices ).6 On the morphological level the kãrakas are represented by
six triplets of case-endings, each of which roughly corresponds with one
kãraka . The kãraka rules are applied with the governing ( adhikãra ) sütra
P. 2. 3. 1 : anabhihite , ( add a case-ending after a lexical unit to convey a
kãraka only) if it is not expressed ( already ). Two of the kãrakas, kartr
and karman , can be expressed by verbal endings, whereas some other kãrakas
can also be expressed by primary and secondary suffixes. The kãraka theory
is of obvious interest to the computer scientist interested in natural language

4 A. Seidenberg, "The ritual origin of geometry, " Archive for History of Exact Sci -
enee , vol. 1, 1962, pp. 488-527.
A. Seidenberg, "The origin of mathematics, " Archive for History of Exact
Science vol 18, 1978, pp. 301-342.
ö S. С. Как, " A frequency analysis of the Indus script", Cryptologia , vol, 12, 1988,
pp. 129-143.
S С. Как, " Indus wrting Mankind Quarterly , vol. 30, 1989, pp. 113-118.
6 * P. Kiparsky and J. F. Staal in Staal, op . cit.
11 Annals, BORI [A. M. ]

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
82 ABO RI : Amrtamahotsava Volume

processing. The reader interested in the details of this th

essays by Joshi7 and Kiparsky and Staal.8

A comprehensive study of A from a computing scie

should include linguistic, structural, and algorithmic asp
must be based on the long tradition of analysis of A tha
2,500 years. Problems of particular interest to the compu
the arrangement of the rules and the smallest set of rules
valent to A. Rearranged rules, such as those by Bhat
Siddhãnata Kaumudf, would provide an invaluable frame
before a comparison can be made from an algorithmic pe
to describe A's rule in a form convenient for analysis by
in mind, we discuss in this introductory paper certain
Pänini's grammar. In particular we consider the followin

1. The sut ra style and the nature of rules.

2, The structure of the rule system.

We will show how the rules can be easily cast in fam

transformational forms. An explicitly algebraic represen
before Pänini's rules can be expressed on a computer so
tional and cognitive implications can be properly assessed

II

The sütra style and the nature of the rules

The sütra style represents a genre of Sanskrit literature. Traditionally

a sütra is defined as the most concise of statements which uses as few letters
as possible. Although many books have been written in the sûtra style»
Pänini's grammar or Pãninisutra , (Ps) is unanimously regarded by tradition
as a model of the sütra style of composition.

Words and sentences constitute the data as well as the rules for
grammar. Language is thus both a means and an end. Pänini's grammar
deals with Sanskrit. But its end language ( the object language ) and the
means language ( the metalanguage ) are distinctly different from each other.
Pänini's matalanguage has its own vocabulary, syntax, and grammar although
it is basically Sanskrit. An extensive use of abbreviated expressions and other
devices has given it an appearance of a code language. It is this feature of
the Pãninisutra that has inspired comparisons with a computer programme. A

7 S. P. Joshi, op. cit . 8 P. Kiparsky and F. Staal, op. cit .

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 Bhate & Как : Panini' s Grammar and Computer Science 83

few prominent aspects of this code language will be described later. A strik-
ing feature of the language of the sütras is the use of abbreviated expressions.
Economy of expression is Pânini's primary concern and he has achieved it by
employing several algebraic devices. Use of technical terms in place of
lengthy expressions is one of them. He uses symbols like ti, ghu , gha and
bha Further, a code representation, technically known as pratyãhãra,
enables him to save words and even letters in a rule. For instance, instead
of directly mentioning the letters y, v, r, l Panini makes use of the pratyãhãra
yan ; for vowels he uses the term ac ; for consonants, hal and so on.

Following is an example of a rule containing all code words :

P. 6. 1. 74 : iko yan aci

i, u, r and I are replaced by y , v, r and / respectively when a vowel

follows.

Use of words without adding endings to them traditionally known as

( avibhaktikanirdeša ) is another striking feature of the language of the Ps .
However, it will not be described here.

IIA . A Ps is a single clause proposition consisting of a subject, a predicate,

and an environment. It is a statement about grammatical features such as a
suffix, an augment, a substitute, accent, reduplication, elision, and compound-
ing. It is usually of the form A is В in the environment С . This can be
written in the following formula :

Ps: A-> B(C)

Here -> stands for is or becomes , and ( ) stands for when , A stan
for the subject, В represents predicate, and С stands for environment. W
A and В are the necessary components of a sütra , С is optional. A uniq
feature of the Ps is the absence of a finite verb predicate. Tradition ho
that the finite verb asti ( is ) or bhavati ( becomes ) is taken to be presen
each rule. A Pãninian rule is thus a statement about something bein
becoming something else. Pânini's marked predilection for nominalizati
reflected in his attempt to reduce all statements to those on being or be
ing. Thus for instance, instead of saying, tat lupyate ,

That ( code letter ) is dropped,

Panini says : tasya lopah (P. 1. 3. 9 ).
Its ( i. e. the code letter's ) elision ( takes place ).

IIB . Another interesting feature of the Pãninian proposition is the

absence of syllogization or any other kind of logical argumentation. Ot

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
84 ÀÈORI : Amrtamalioisava Voíumê

sUtra works such as the Nyayasütra and the Brahmasütra

addition to A, B, and C, a cause ( hetu) and an illustratio
his grammar Panini never poses the question, Why? His
ments of linguistic facts in reply to the question, What ? In
describes facts of language without accounting for them. T
language of the Ps consists of three types A, B, and С as sh

IIC. The relation between A and В oil the one hand and that of С with A
and В on the other is expressed by Panini by the use of certain cases. While
the predicate item is always used in the nominative, the case of the subject
item is determined by its specific relation with the predicate. For instance, if
the predicate is a substitute ( adeša ) the subject is used in the genitive. This
has been directly stated by Panini in a rule.10 If on the other hand, the
predicate is a suffi ' (pratyaya), the subject is. put in the ablative.11 Envi-
ronment is expressed in the locative if it follows the subject.12 These state-
ments can be put in the following formula :

For substitute : A ( gen ) -> В ( nom ) ( С (loe) )

For suffix : A (abl ) - > В ( пот ) (С (loe))
A few more formulas can be formed on the basis of other observations. ia
The following rule is an illustration of the first formula :

P. 6. 1. 74 : iko yan aci

yan is substituted in place of ik when ac follows.

The genitive form ikah, the nominative form yan and the locative form ac
are in accordance with the statement made above and the mutual relationship
among the three items is conveyed by the case-endings.

HD. Science may be viewed as a body of generalizations followed by state-

ments of exception when necessary. Päninian science of grammar also con-
sists of a set of general rules followed by statements of exception. A Ps
can be thus either a generalized statement or a specific statement which stands
as an exception to a generalization. In a generalized as well as particular
statement the subject or the predicate can be a multi-member category. A
single predicate may be shared by many or all subjects. For instance consider

9 It must be borne in mind that exceptions to such a structure for Ps do exist. But these
exceptions do not define the general structure of the rules.
10 P. 1.1.49. 11 P. 1.1.67.
12 P. 1.1.66.
*3 For details see Saroja Bhate, 44 Some aspects of Panini's sntra style of compos
B. R. Modak Felicitation Volume , March 1989, pp. 37-46.

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 Bhate & Как : Pàttini' s Grammar and Computer Science 85

P. 3. 2. 1 : ( dhãtoh ) kar many an

The suffix an is added to a root in the sense of object.

The word dhãtoh which is put in the bracket ( the reason will be explained
later) is the subject and an is the predicate. Here dhãtu stands for any root
in general. The statement of the suffix an thus is applicable to all roots in
general. The predicate an is thus shared by all subjects. Here the subject
which is a multi-member category, is represented by a class term ( i. e.
dhãtu ). This type of Ps can be represented as :

On the other hand, sometimes many predicates are linked with one subject.
For example,
P. 5. 2. 32 : ner bidacbirîsacau

P. 5. 3. 33 : ( ner )inacpitaccikaci ca
The suffixes bidac , birïsac , inac and pitac are added to ni (in the sense
of flat nose ) and in case of the later two suffixes ni is replaced by cika
and ci respectively.

Here as many as four predicates are shared by one single subject, ni.
This statement could be represented as :

A -> В1-" ( С)
A third type in which both A and В are simultaneously multi-member catego-
ries is also occasionally met with. For instance,
P. 3. 1. 133 : ( dhãtoh ) nvultrcau
The suffixes nvul and tre are added to any root.
This statement is of the type

A1~n->B1~n{C)
Just as is true for A and В, С also can be a multi-member category. For
instance,

P. 1. 3. 13 : ( dhãtoh ) bhãvákarmanoh ( ãtmanepadam )

A tmanepada endings are added (to a root) in' the sense of bhãva
( state ) or karman ( object ).

Bhãvakarmanoh expresses the environment in terms of meaning. The

two meanings bhãva and karman are mentioned here. This can be repre-
sented as.

Ач-В(С1~п)

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
86 ABORÌ : Amrtamahotsava Volume

Finally, combination of all the three multi-member categories is als

met with in certain sütras. For instance,

P. 3. 4. 70 : ( dhãtoh ) tayor eva krtyaktakhalarthãh

The suffixes krtya, kta and those conveying the same meaning as th
of khal are added to any root in the sense of bhãva ( state ) or karm
(object).
The term tayoh is to be interpreted as bhãvakarmanoh. We thus have a
statement of the type

Al -rt ^Bi-n(Cl-n)
Observations made above hold true of both the general and special rules in
Panini's grammar. To sum up, the three categories А, В and С may be
either single-member or multi-member categories. They appear in all permu-
tations and combinations in Ps.

1IE. Now it remains to be seen whether Panini has provided any clarifica-
tion regarding the application of the multi-member categories. Two questions
arise when a statement contains multi-member categories :

1. Are the members in a category linked to each other disjunctively or con-

junctively ?
2. What are the mutual relations between the members of two multi-member
categories ?

Question 1 :

Panini has employed three linking devices in the A, namely, juxtaposition,

dvandva compound, and the particle ca. They link either items or statements.
We are at present concerned with linking of items. These devices work in
terms of disjunction or conjunction. Disjunction implies application of all
the items separately whereas conjunction implies their application together.
Items put together in a dvandva compound are disjunctively connected. For
instance,

P. 3. 1. 133 : ( dhãtoh ) nvultrcau

The suffixes nvul and tre are added to any root.

Here the compound nvultrcau is a multi-member predicate. The items n vul

and tre are disjunctively connected with each other. Therefore they are sepa-
rately and not simultaneously added to a root. Thus we can derive two
separate forms such as pãcaka and paktr from root рас . On the contrary, if

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 Bhate & Как : Panini* s Grammar and Computer Science 87

the items are put in juxtaposition they are conjunctively connected with each
other and are, therefore, simultaneously applicable. For instance,
P. 7. 4. 49 : sah syãrâhadhãtuke ( tah )
s is replaced by t when an ãrdhadhãtuka ending beginning with s
follows.

Here two items si and ärdhadhätuke which belong to the category С are not
put together in a compound, but are mentioned separately in juxtaposition.
Therefore they are conjunctively connected. In other words si and ärdha-
dhätuke are co-referential. Whenever items belonging to one category are put
in juxtaposition in a rule they hold a head-modifier ( or adjective-substantive )
relationship. Two or more items belonging to the same category and yet not
connected by a head -modifier relation never occur in juxtaposition in a single
sutra . Juxtaposed occurrence of two heads or modifiers always indicates the
existence of separate sUtras . The particle ca is never used to link two or
more items as it does in ordinary language.14 Items belonging to the same
category in a rule are either put in a compound or are juxtaposed according
to their relation with each other.15

Question 2 :
Panini accepts the principle of numerical correspondence for linking items in
two multi-member categories. He states the rule as follows :

P. 1.3. 10 : yathãsankhyam anudešah samãnãm

Items ( in two categories ) having the same ( number ) are connected
( with each other ) in their respective number ( i. e. order ).

Take, for instance, P. 4.3.94: tãdiÉalãturavarmmatikucavãrãd dhakchandha -

ñyakah
The suffixes dhak9 chart , dhañ9 yak are respectively added to ( the
stems ) tüdí , salatura , varmmatï , and kãcavãra ( in the sense ' it is the
place where his ancestors lived ' ).
Here both A and В consist of four members each and the members of A are
connected with the members of В in the same order in which they are put in
the compound.

14 For example, ramas ca kr<¡nas ca gacchatah. Here ca links two items disjunctively.
The particle ca can also link two or more items conjunctively in ordinary language ; for
example ' He eats curds and honey '
15 For details see S. D. Joshi and Saroja Bhate, The Role of the particle ca in the
interpretation of the Astãdhyãyi. Publication of the Centre of Advanced Study of
Sanskrit, University of Poona, 1983.

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
8.8 ABORI : Amrtamahotsava Volume

III

Arrangement of the Rules

As stated earlier a PSninian rule consists of three elements : A, B, and

C, the last being optional. All these elements are not always explicitly present
in the wording of a sût ra. Just as a finite verb form is implicit, a certain
element is understood to be present in a sãtra from the context. While inter-
preting certain rules commentators actually borrow the missing term from the
preceding rule. This borrowing or continuation of a word or words is techni-
cally called anuvrtti. The procedure of anuvrtti is nothing but ellipsis which
is a regular feature of ordinary language. While ellipsis is optional and has
an ad hoc character in ordinary language, anuvrtti is a systematic and mecha-
nical device in A . The sutras in A are arranged in such a manner that a rule
borrows an item or items from the preceding context. By putting together
such rules which share an item or items in common Pãçini has been able to
achieve economy of expression to a large extent. A few examples of anuvrtti
will show the working of this device :

P. 1. 4. 14 : suptiňantam padam
That which ends in sup or tin endings is called pada.
P. 1. 4. 15 : nah kye ( padam )
That which ends in n ( is called pada when the suffix kya follows. )

The predicate item padam which is missing in P. 1. 4. 15 is continued from

P. 1.4. 14. ( The missing item when borrowed from the preceding rule is put
into brackets ).

P. 3. 3.114: napumsake bhãve ktah ( dhãtoh )

The suffix kta is added ( to the root in the sense of bhãva, state, and
the form is used in neuter ).

P. 3. 3. 115 : lyut ca ( napumsake dhãtoh )

And suffix lyut ( is added to a root in the sense of bhãva and the form
is used in the neuter ).

The environment expressed by the terms napumsake and bhãve is carried for-
ward in P. 3. 3. 115 from P. 3. 3. 114. The subject item dhãtoh which is
carried forward in P. 3. 3. 114 is also continued in a number of rules. For
example, the term dhãtoh which is mentioned- in P. 3. 1.92 is continued
throughout the following chapter, nearly in 500 rules. Through the device of
çnuvrtti Panini has been thus able to avoid repetition of the word dhãtoh in

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 Bhate & Как [ Panini* s Grammar and Computer Science 89'

more than 500 rules. Anuvrtti is thus intrinsic with the style of theP,y.
Although Panini has arranged rules in his grammar mainly on thematic basis,
the arrangement of rules within different sections is totally governed by the
dictates of anuvrtti. A very important difference between anuvrtti and ellipsis
inordinary language consists in the fact that while the latter is dependent
upon expectancy and the listener's ( or rather receiver's ) intention, the for-
mer is obligatory. Items in the previous rules must continue in the subse-
quent rules. Expectancy is not just sufficient ground for continuing an item.
An item is found to be continued even when there is no expectancy. For
example,

P. 1. 2. 4 : sãrvadhãtukam apit
A sãrvadhãtuka suffix, other than the one which is pit , is nit,

P. 1. 2. 5 : asamyogãl lit kit ( apit )

A lit suffix other than the one which is pit added to ( a root ) not end-
ing in a conjunct consonant is kit .

All the three elements, namely subject ( lit ), predicate ( kit ) and environment
( asamyogãt ) are present in P. 1. 2. 5. It presents no expectancy for any item
in order to complete its meaning. Yet the item apit is continued in the rule.

There are, however, some constraints on this flow of anuvrtti . The

major constraint is that an item is carried forward in the subsequent rules
until it is blocked by an incompatible item. Thus compatibility and incom-
pability play a major role in deciding anuvrtti of words. For instance, in the
example given above the item nit which is continued in P. 1. 2. 4 is not further
continued in P. 1. 2. 5 because it contains the item kit which is incompatible
with nit. Two incompatible items do not exist in a rule except under some
special circumstances. The fundamental rule of anuvrtti can thus be stated
as follows :

An item is continued in the subsequent rules unless it is blocked by an

incompatible item.

Two items are incompatible if they belong to the same category ( i. e.

subject, predicate, or environment ).

Again in the same example quoted above the items sãrvadhãtuka and lit are
incompatible with each other. Therefore, the former is not continued in
P. 1. 2. 5 as it is blocked by the latter.

Items which are incompatible with each other usually appear in the same
case-ending. However, appearance in the same casç-ending is not the only
'2 Annals» BORI I A. M. )

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
90 ABORI : Amrtamahotsava Volume

identification mark of incompatible items. Their relative sy

has also to be taken into consideration. Turning back aga
example, the two items sãrvadhãtukam and apit together fo
category in P. 1.2.4. While sãrvadhãtukam is the head i
modifier ( adjective ). There is obviously no incompatibility
and a modifier. This is true not only when they belong to on
rule as in the above case, but also when they are mentioned
rules. Thus modifier item apit mentioned in P. 1. 2. 4 is com
head item in the subsequent rule. Therefore, although the h
P. 1.2. 5 blocks the incompatible item sãrvadhãtukam in
rule, it does not block the modifier item apit, which is ther
in P. 1. 2. 5. Another rule of anuvrtti may be laid down on t
observation as follows :

A head item blocks an incompatible head item, but it does not block
a modifier if it is not incompatible.

A modifier blocks an incompatible modifier, but it does not block a

head item if it is not incompatible.

There are, however, cases when a modifier is not continued since it is

incompatible not on syntactic, but on semantic grounds. ( These cases will
not be discussed here as they have no direct bearing on the present topic )

Arrangement of the sütras in the A is initially topicwise. Thus the Ps

begins with definitions of various technical terms and rules of interpretation
and treats various types of derivations such as compounds, primary deriva-
tives and secondary derivatives in separate sections. The last part of the A
is devoted to morphophonemics including euphonic combination, accent and
tone. Within a thematic group the sutras are arranged on the basis of the
principles of anuvrtti . Although a generalization is followed by specific or
individual rules, this order is often violated due to exigencies of anuvrtti .
Anuvrtti is thus a key- word for the arrangement of the Л.16

IV

Techniques of the Description

In addition to anuvrtti and artificial terminology including pratyãhãras

Pãnini employs the device called anubandha. An anubandha is a code-letter
which indicates a grammatical function like elision and reduplication. Pãçini

16 For a discussion of the principles of anuvrtti see S. D. Joshi and Saroja Bhate, The
Fundamentals of Anuvrtti , Publication of the Centre of Advanced Study of Sanskrit,
University of Poona, 1984,

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 Bhate & Как : PãninVs Grammar and Computer Science 91

has made use of almost all vowels and consonants as symbols for various
functions. Anubandhas are added to various grammatical units such as
suffix, an augment and a root. For example, the suffix a is mentioned as an
where the code letter n suggests that the vowel ( either initial or final depend-
ing upon the type of derivation ) of the stem to which the affix is added is
lengthened.17 The anubandha к attached to an augment indicates that it is
added at the end of an element.18 Thus the augment t which is mentioned as
tuk in the rule hrasvasya piti krti tuk ( P. 6. 1. 69 ) is added after an element,
e. g. in the form adrtya it appears after root dr which ends in a short vowel.
The anubandha n attached to a verbal root indicates that the root is conjuga-
ted in the middle voice. Anubandha is thus a powerful device.

IVA . A major aspect of Panini's descriptive technique is the law of utsarga

and apavada that relates exceptions and individual rules. Although Pacini
never explicitly states the law of utsarga and apavada it is part of the inter-
pretative apparatus used with the Ps .19 The law of utsarga and apavada
states that an apavada ť exception ' is more powerful than an ustarga
' general rule '. Therefore before applying the utsarga one has to give check
for its apavada(s). The ustarga thus occupies the area not occupied by its
exceptions. Further, once an utsarga is barred from entering into the area
of its exception, it can never enter the area again. For example :

P. 4. 1. 92 : tasyãpatyam ( an )R

( The suffix an is added to a noun in the sense ) ' his offspring. '

P. 4. 1. 95 : ata iñ tasyãpatyam

The suffix iñ added to a noun ending in a ( in the sense « his offspring ' ).
P. 4. 1. 95 is an exception to P. 4. 1. 92. Therefore, the suffix an taught by
P. 4. 1 . 92 is barred from being applied to stems ending in a . Thus from the
stem daksa is derived the patronymic dâksî ( daksa + iñ) and not daksa
+ art.

Sometimes application of the utsarga even in the domain of apavada is

desired. In such cases Panini announces that the apavada operates optionally #
For instance,

P. 4. 1. 121 : dvyacah ( dhak striyãh )

» P. 7. 2. 115 to 117. is P. 1.1.46.

18 For example, the rules of interpretation ( paribhãsã ) 57, 58, 59, 60, 62, 64 a
the Paribhãsendusekhara of Nãgojibhatta edited by K. V. Abhayankar, Bhan
Oriental Research Institute, Pune 1962.

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
92 ABORI : Amrtamahotsava Volume

( The suffix dhak is added to a feminine noun ) con

vowels ( in the sense ' his offspring ' ).
P. 4. 1. 118 : pílãyã vã ( an striyãh)
( The suffix an is added ) optionally to pila ( in the sense 'h

The option marker vã in P. 4. 1. 118 suggests that the excep

an operates optionally. Therefore, the utsarga suffix dhak t
121 is also applied and two alternate forms, paileya ( pî la
( pila + an) are derived.

IVB . An extremely important principle is the siddha princi

Panini does not directly mention it, his statement of the
( P. 8. 2. 1 ) implies it. Traditionally the whole A is divided
the basis of P. 8. 2. 1 : ( 1 ) the siddhakãnda ( P. 1.1.1 to the
section of the eighth Chapter ) and ( 2 ) the asiddhakãnda or
to the end of the fourth section of the eighth Chapter ). Tr
the adhikãr a, ' chapter heading 8.2.1: pUrvatrãsiddham
every rule is regarded as ) not having taken effect with referen
ones '

The term siddhakãnda implies that any rule in this par

« having taken effect ' for any other rule in the whole of A
before being effective a rule takes into consideration possibi
of other rules. The sequence of rules in the book does n
derivational process. What matters is the siddha relation a
The finite verb form bhavati is, for instance, derived as follows

bhu + lat P. 3. 2. 123

bhu + tip P. 3. 4. 78
bhu + sap + îi P. 3 1. 68
bho + a+ti P. 7. 2. 115 ; P. 6. 1. 78
bhavati

It will be clear from the derivational stages given abov

in the first section of the third Chapter applies after the r
section of the same chapter and the rule in the sixth Chapte
rule in the seventh Chapter. These rules are siddha for each o
can feed each other ( the application of P. 6. 1 . 78 is depende
tion of P. 7. 2. i 15 in the present example ). This free move
all directions is implied by the siddha principle. Yet this arb
of rules within 1. 1-8. 1 is restricted somewhat by a category

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
 Ëhate & Как : Panini9 s Grammar and Computer Science 9Ì

ordered pairs. In each pair the rule that is applied first is called antaraňga
and the rule that is applied next is called bahirariga.

On the contrary the rules in the asiddhakãnda are operative only in

one direction.

P. 8. 2. 1 : pUrvatrãsiddham

states that all the rules stated subsequently are asiddha, and effective for the
rules stated earlier, that is for the rules in the siddhakãnda. Similarly for
each rule in the asiddhakãnda all subsequent rules are asiddha. In other
word, rules in the asiddhakãnda operate in the same order in which they are
arranged. For example, the form pakva is derived from root рас as follows :

рас + ta P. 3. 2. 102
рак + ta P. 8. 2. 30
рак -f va P. 8. 2. 52
pakva

It is clear from the procedure given above that P. 8. 2. 30 precedes the

application of P. 8. 2. 52. In fact, if P. 8. 2. 52 is applied first P. 8. 2. 30 cannot
be applied since the environment favorable for its application does not exist.
The rules in the asiddhakãnda must therefore apply in the same sequence in
which they are stated by Panini.

Both the siddha and asiddha principles have been recently studied
carefully leading to important new insights.20

Concluding Remarks

Our analysis was meant to highlight several formal features of Pânini's

grammar that have direct parallels in computer science. What might be other
features of the grammar that have not yet been rediscovered in computer
science remains to be seen. But the very success of A suggests that aspects of
its structure will have implications for further advance in computer science,
knowledge representation, and linguistics. In particular we can hope for
significant applications in natural language processing. The ongoing analysis
of the structures of Panini and those of the later grammarians and logicians

20 S.D. Joshi and Paul Kiparsky, " Siddha and asiddha in Pãninian phonology,'
in Current Approaches to Phonological Theory , edited by Ď. Dinneen. Bloom-
ington, 1979.

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms
94 AÉOŘI : Amrtamakotsava Votume

will be aided by the development of software to implement

computer.

The specific issues of immediate interest to the computer scientist in-

clude analysis of the arrangement of the rules and search for other arrange-
ments that are equivalent in terms of their generative power. The formal
aspects of these arrangements and their relationships is likely to help define
the notion of distance between grammars. Such a notion is of immediate
relevance for machine translation. Given two languages with grammars that
are close in structure, as in the Indo-Aryan family of languages, one would
expect the translation across the languages to be relatively easy. A formali-
zation of the notion of closeness is also likely to give pointers regarding how
an automatic translation might proceed.

One great virtue of the Päninian system is that it operates at the level
of roots and suffixes defining a deeper level of analysis than afforded by recent
approaches like generalized phrase structure grammars21 that have been in-
spired by development of computer parsing techniques. This allows for one
to include parts of the lexicon in the definition of the grammatical structure.
Closeness between languages that share a great deal of a lexicon will thus be
represented better using a Pãninian structure.

These fundamental investigations that have bearing on linguistics, know-

ledge representation, and natural language processing by computer require
collaboration between computer scientists and Sanskritists. Computer orien-
ted studies on A would also help to introduce AI [ artificial intelligence ], logic,
and cognitive science as additional areas of study in the Sanskrit departments
of Indian universities. This would allow the Sanskrit departments to comple-
ment the programme of the computer science departments. With the incor-
poration of these additional areas, a graduate of Sanskrit could hope to make
useful contributions to the computer software industry as well, particularly in
the fields of natural language processing and artificial intelligence.

21 G. Gazdar, E. Klien, G. Pullum, I. Sag, Generalized Phrase Structure Grammar .

Cambridge : Harvard University Press, 1985.
Gazdar et al . claim that context-sensitive rules can be replaced by a larger set of
context-free rules for all natural languages, excepting Bambara ( Mande family, West
Africa ) and a certain Swiss dialect of German. Context-free rules make parsing by
computer easier.

This content downloaded from

203.110.242.40 on Thu, 05 May 2022 07:01:13 UTC
All use subject to https://about.jstor.org/terms