On the alleged descriptive meaning of Pāṇini’s Aṣṭādhyāyī

J. F. Staal in his seminal paper “Euclid and Panini” published in 1965, claims that, since:

(i) “Linguistics is not prescriptive, but descriptive”, and,

(ii) Panini’s rules are exclusively derived from ordinary usage”, then

(iii) the principal problem A is concerned with is “to give and adequate description of
ordinary usage”1. The claim expressed by the conclusion (iii) has become a widely common
accepted assumption, which has been repeated, virtually without much dispute, by just about any
major commentary work on Pāṇini’s A.

To quote just but some few examples, Cardona reproduces literally Staal’s claim2,
although he later dismisses later Staal’s even stronger claim as to considering P a
transformationalist, on the ground that he “does not derive one sentence from another, nor does
he operate with actual embedding”3.

S. D. Joshi, for his part, emphasizes that:

“Pāṇini’s descriptive study of the Sanskrit language, according to Patañjali, is primarily

concerned with the formation of the significant classes of words … His purpose is to describe the
mechanical procedure of the word-formation ,by the skilful application of grammatical rules, an
act which is considered by modern linguistics as highly creative”4.

Harold G.Coward and K.Kunnjunni Raja acknowledge that:

“Pāṇini’a basic work is merely titled “The Eight-Chaptered” (Aṣṭādhyāyī) … providing a model
for recent and contemporary work in descriptive linguistics than can stand with the best efforts of
modern analysts. The eight chapters constitute a complete descriptive analytical grammar of the
Sanskrit language”5.

1
Staal 1965,109
2
Cardona 1980, 182..
3
Ibid., 236.
4
Joshi 1965/2005, 52

5
Coward & Raja 1990, 14.

1
 Based on the historical evidence that has come down to us, the discussion about the
meaning of A’s grammatical rules, whether they are descriptive or prescriptive was never an
issue amongst the Ancient Indian grammarians. In that sense, the problem this paper is
concerned with is undecided, in so far as there is no direct evidence bearing on a particular
ancient interpretation, which may be ultimately used to argue in one way or another. This lack of
historical evidence has given much weight to the prevalent interpretation of Pāṇini’s A,
according to which, the rules of the A -without much further qualification- are to be regarded as
descriptive rules.

While it is a fact that P’s rules are, as Staal puts it, exclusively derived from ordinary
usage, it does not necessary follow that they are descriptive, unless the inference is supplemented
with the additional premise that P’s grammatical rules may be correctly interpreted as an
example of generative grammar rules6. The premise is provided by Joshi and Roodbergen. “The
A, so the claims goes, is “a generative grammar”, and they spell out subsequently the two senses
in which the term “generative grammar” is likely to be taken:

(a) in a Chomskyan sense, for A describes a process by means of which the

“derivation of word form is fully described”, and
(b) in the sense that, with the help of a limited number of rules (about 4000), the A
…is able to produce an infinite number of words and thus an infinite number of
sentences”7.

From claim (a) it follows that Pāṇini’s A aims to provide a particular set of formal rules
which describe the formations of words and sentences in Sanskrit. From claim (b) it follows that
the A may be correctly described as a computational mechanism capable of recursively
generating the set of words and sentences in Sanskrit. Both claims (a) and (b) press hard for
interpreting the grammatical rules of the A as general hypothetical statements, which based on a
corpus of the linguistic data provided by the language (bhāṣā) of the śiṣṭas, may be said to
describe correctly the e-language ‘Sanskrit’ as used by the śiṣṭas. Part of the evidence to sustain
claims (a) and (b) come from Chomsky’s own recognition that “Pāṇini’s grammar can be
interpreted as a fragment of such a “generative grammar” in essentially the contemporary sense
of this term”8.

I should like to examine, in particular, the evidence that has been provided to justify the
descriptive character of P’s rules on the ground that the A can be meaningfully regarded as a

6
See on that Cardona 1990, 232.
7
Joshi & Roodbergen 1990, 15‐6.
8
Chomsky 1969 ,v.

2
generative grammar in a Chomskyan sense. But before doing that, I shall try to remove some
misunderstandings, or so they seem to me, related first to the P’s conception of metalenguage,
and secondly to the alleged formal properties that, according to both Staal and Joshi, the A’s
language may be accounted for.

II

Scharfe claims that the description of P’s technical terms proves that “Panini himself recognized
two distinct language systems”9. According to Scharfe, one language system is Sanskrit, while
the other -the metalanguage- is the language P used to state the sūtras. On this view Scharfe
concludes that the metalanguage used by P does not depend from Sanskrit and it is artificial

Now, Scharfe’s claim seems to be based on an unstated presupposition according to

which the distinction between language and metalanguage is likely to become ambiguous, if it
does not map somehow the distinction normally made in formal language between object
language and metalanguage, where metarules are stated in the metalanguage.

The statement of metarules involves the use of metavariables, e.g., variables referring to
other variables which have a different content from those belonging to the objet language. For
instance, the statement:

(1) p ∨ ¬p

is a statement in the object language, as such it can take a truth-value, in this case being a
tautology it will be true for all possible interpretations. Statements of the objet language need to
be interpreted in order for them to have a meaning, while metarules state the formation rules
according to which the formulae of the object language are formed. Take, for instance, the
following metarule:

(2) A, B

_________

A&B

It states that if a formula A is a theorem and the formula B is also a theorem, that is to say if both
A and B have derived in the language, either from the set of axioms, or from other propositions
using inference rules, then the conjunction of A and B “(A & B)” is also a theorem.

The metarule expressed in (2) is not a statement, and consequently it cannot be assigned
any truth-value. Logical rules are expressed in the metalanguage. The expressions “A” and “B”

9
Scharfe 1971, 4.

3
stand for any statement and it is just lies down an inference rules. Now, inference rules are
neither true nor false, they are just sound or unsound rules. They either preserve or do not
preserve the truth-value of first order statements. There are therefore some structural properties
that first order statements do not share with metarules, and those properties have to be shown in
the syntax of the language. By using metavariables the distinction between first order statement
and inference rules is assured, in order to avoid any truth assignment on inference rules.

It is not necessary for descriptive grammatical purposes, however, to recur to the notion
of formal language in order to make the distinction between language and metalanguage explicit.
Since just about any natural language may likely be used as metalanguage of its own. Indeed, it
perfectly possible to define several levels of metalanguages using any natural language, as shown
in what seems to be a bewildering conversation between the White Knight and Alice, from a most
quoted passage of Lewis Carrol’s Through the Looking-Glass:

“‘You are sad,’ the Knight said in an anxious tone: ‘let me sing you a song to comfort you.’
‘Is it very long?’ Alice asked, for she had heard a good deal of poetry that day.
‘It’s long,’ said the Knight, ‘but very, very beautiful.
Everybody that hears me sing it— either it brings the tears into their eyes, or else— ’
‘Or else what?’ said Alice, for the Knight had made a sudden pause.
‘Or else it doesn’t, you know. The name of the song is called “Haddock’s Eyes.”’
‘Oh, that’s the name of the song, is it?’Alice said, trying to feel interested.
‘No, you don’t understand,’ the Knight said, looking a little vexed. ‘That’s what the name is
called. The name really is “The Aged Aged Man.”’
‘Then I ought to have said “That’s what the song is called”?’ Alice corrected herself.
‘No, you oughtn’t: that’s quite another thing! The Song is called “Ways and Means”: but that’s
only what it’s called, you know!
‘Well, what is the song, then?’ said Alice, who was by this time completely bewildered.
‘I was coming to that,’ the Knight said. ‘The song really is “A-sitting On A Gate”: and the tune’s
my own invention’”.

Alice’s vexing perplexity is largely due to the White Knight’s remarkable ability to
introduce up to four different levels of reference, without resorting to the formal notion of
metalanguage. Using the object-language, he states that the song is “A-sitting On A Gate”. Next,
he informs that the name of the song is “Ways and Means”, this is the first level of
metalanguage, where the name “Ways and Means” is used to refer to the name “A-sitting On A
Gate”. He goes onto explaining what the song name’s name is (i.e., how the song is actually
called), which is “The Aged Aged Man”. Now, the name “The Aged Aged Man” defines the
second metalinguistic level, where names may be used to refer to names for other names. Finally,
the third level of metalanguage is introduced when he explains that “Haddock’s Eyes” is the
name of how the song is called.

4
 There are no logical reasons to think that hierarchy of names and metalanguages may go
as high as necessary, but that is not the moral to be drawn from Lewis Carroll’s example. What I
think it shows is that you can have as many different metalinguistic levels as you need to ascribe
different properties to different objects, if adequate measures are taken to clearly distinguish
reference levels. The A states a rule that, despite the wide scholarly discussion it has generated
may be considered as a general metarule governing the use of different metalinguistic levels:

1.1.68 svaṃ rūpaṃ śabdasya aśabdasaṃjñā [(when a metalinguistic item is mentioned in

a rule for purposes of grammatical operations, then) the own (phonetic) form of the
metalinguistic item (is to be understood), with the exception of the technical name for the
metalinguistic item”.]10

The rule is meant to assure that words used in the A should be taken as a name of their own,
except when a technical term occurs, in which case it is used with the meaning it has been
assigned. From rule 1.1.68, which is stated in the second level metalanguage, it follows that all
rules defined by the A are formulated in the metalanguage, except in those cases where a
technical definition is introduced, in which case it stated in the object language where a
particular saṃjñin is assigned the corresponding saṃjñā.
Scharfe is committed to declaring that, though P did not have in mind the formal
distinction between language and metalanguage, he should have kept, nonetheless, a different
assignment of structural properties for the statements of the metalanguage, and consequently in
rules:

1.1.49 ṣaṣṭhī sthāneyogā [(a word ending in) the sixth case ending is to be connected with
(the word) sthāne ‘in the place”. ]

1.1.66 tasminn iti nirdiṣṭe pūrvasya [when (a grammatical item) has been mentioned as
locative form, (then the grammatical operation is to be applied) to (the item) immediately
preceding (the item mentioned in the locative).]

1.1.67 tasmād ity ūttarasya [when (a grammatical item) has been mentioned as a ablative
form, (then the grammatical operation is to be applied) to (the item) immediately following (the
item mentioned in the ablative)]

where the genitive, locative and ablative cases are meant to be used metalinguistically, “[t]he
meanings of the genitive, ablative and locative, through derived in some way from those these

10
English translations of the A, unless otherwise stated, have been taken from Joshi & Roodbergen 1991.

5
cases have in the object language, are so special and so technical, that new definitions were
necessary”11.

Scharfe’s concern is that P should have not used the expressions ṣaṣṭhī, tasminn and
tasmād as a names mentioning the grammatical technical terms related to each respective case
ending. Yet he fails to realize that technical items may be used metalinguistically. Similar cases
may be found in: 1.4.105, 1.4.106, 3.3.143, 7.2.103 and 7.4.106. For Scharfe “[t]hese few non-
technical cases of metalinguistic terms are best regarded as slips violating the style and system of
the metalanguage”12.

The truth is, however, that those cases do not violate the style and system of the
metalanguage anymore than the statements made by the White Knight may violate the style and
the system of English, when used as a metalanguage.

III

Staal, who in a later paper called Panini, the “Indian Euclid”, draws a parallelism between
Euclid’s Elements and the sutras of the A “The sūtras”, accordingly, “can be divided into three
types in a way roughly paralleled to Euclid’s axiomatic system:

(3) The sūtras proper, or theorems (vidhi), which describe linguistic facts while
emphasizing word formation.

(4) The defining sūtras, (saṃjñā sūtra) which introduce technical terms, eg.,
“homogeneous” or “”homorganic” (savarṇa) for sounds which are pronounced in the
same place and with the same tension of the mouth. Finally,

(5) the metatheorems (paribhāṣā sūtra), which explains how rules have to be treated and
applied in particular cases”13.

This claim was initially stated in an earlier paper published in 1961 where he also announced the
descriptive nature of the A:

“Pāṇini’s grammar (fourth century B.C) owes part of its fame to the fact that the rules
in it do not prescribe how to speak correct Sanskrit, but describe the facts of the
language”14.

11
Scharfe 1971, 32.
12
Scharfe 1971 ,34.
13
Staal 1965, 109.
14
Staal 1961,122.

6
 When he comes down in dealing with technical terms, he notices that:

“technical terms belong to the metalanguage, otherwise consisting of the rules

paribhāṣā which are rules of interpretation or rules which indicate how rules of
grammar have to be manipulated… The status of a paribhāṣā corresponds to the
status of a metatheorem in modern logic”15.

Although the comparison Staal draws between theorems and vidhi was initially made in
reference to Euclid’s axiomatic system, the subsequent analogy between metatheorems of
modern logic and paribhāṣā rules seems to be based on the assumption that there is some sort of
theoretical correlation between the language of first-order logic and generative grammar. To see
this we may view the language of first-order logic, when considered as an axiomatic system, as a
set of rules that recursively defines an infinite set of strings. The set of the strings, i.e., the set of
propositions derivable in the system, may be said to have been generated by the axioms of the
language together with the rules of inference.
Equally, we may think of generative grammar, at least in its minimalist version, as a set
of axioms -in this case one single axiom, called “merge”- that together with the set of phrase-
structure rules and the transformational rules for a natural language, can generate a set of
structural descriptions, in addition to a set of string of morphemes and words, the strings being
the last lines of the structural descriptions.
Having said that, we can see that a particular correspondence stands between first-order
logic and generative grammar: the set of phrase-structure rules and transformational rules for a
natural language would correspond to the rules of inference in the language of first-order logic,
while the a set of structural descriptions and the set of strings of morphemes and words would
match respectively the proofs and the theorems of the calculus.
These correspondences may be used then to check if Staal’s analogies are acceptable.
Clearly, he understands vidhi rules in terms of theorems. The analogy, consequently fails at this
point. Strictly speaking, theorems should rather be equated with the set of strings of morphemes
(pratyaya affixes) and words (pādas) and not with vidhi rules, that generally prescribe or prohibit
a particular grammatical operation and should be equated more properly with transformational
rules. Rule:
1.4.14 suptiṅantaṃ padam

sets the definition of pāda as that which ends in suP and tiṄ affixes. It shows that padas are the
result of a consistent application of rules governing the mapping of affixes suP and tiṄ onto
verbal and nominal roots, according to the division introduced by rules 4.1.2 and 3.4.78
respectively. Vidhi rules including pratyaya, and for that purpose the rest of operational rules
(āgama, ādeśa, vikāra and lopa) are never proven in the A, and if consider as transformational

15
Staal 1961,123.

7
rules, they should rather be equated with the set of rules of inferences in the first-order logic,
which Staal mistakes for paribhāṣā rules.

Staal’s confusion is due to the peculiar role paribhāṣā rules play in the A, being rules
about rules they are prone to be confused with the metarules of first-order logic, that Staal
regards as “metatheorems”.

IV

On the same bend as Staal’s claims, S. D. Joshi has emphasized that: “Panini’s system is closely
connected with modern logical principles in the technique of ‘description’. In analyzing the
Sanskrit language, Panini follows the principles of mathematical calculus”16. Joshi’s proposal
rests on a particular way of interpreting the analytic method (yathoddeśapakṣa) and the synthetic
method (kāryakālapakṣa) in understanding the meaning of P’s rules. Accordingly, the method
yathoddeśapakṣa, ‘gives us the axioms or the algebraic formulae of Panini, while
kāryakālapakṣa “involves the application of the mathematical calculus in generating a new chain
out of various former chains”17.

What he calls the modern logical principles in the technique of ‘description’ becomes
apparent later, when he specifies under the heading Formation-specification, the string of
symbols which presumably are counted as the well formed formulae (WFF) of P’s language. He
lists in (i)-(v) the set of primitive items, roughly corresponding to the notion of terms in first
order logic and formation rules for WFFs.

(i) primitive signs as ̎technical devices for forming the morphemes from WWFs”

(ii) WFF derived from Gaṇapāṭha and suffixes.

Two formation rules:

(iii) Concatenation rule, and

(iv) The formation rule for deriving verbs from verbal roots and suffixes.

And finally (v) as a substitution rule18. It seems that Joshi is just mapping onto the A’s structure
the familiar set of recursive definitions as given in the first order logic language. As seen above,
a set of rules that recursively defines an infinite set of objects may be said to generate this set. In
first order logic language, the WFFs are generated by recursive definitions, but P does not follow
the same structural pattern in the A.
1616
Joshi 1965/2005,56.
171717
Joshi 1965/2005, 57.
18
Joshi 1965/2005,58.

8
 Some Joshi’s strong claims need to be clarified in order to avoid some misunderstanding.
Clearly, P is not following the principles of mathematical calculus in the A, which were totally
unknown at the time P composed the grammar. Consequently, it cannot be maintained as a
correct historical attribution, on the pain of incurring in a severe misconception, that P “presents
such an axiomatic system that every stage of grammatical structure is obtained from the WFFs
and the rules of substitution and generation”19.

Second, the A was not conceived in any imaginable sense as anything remotely close to
an axiomatic system. That, however, is not to deny that the A has to secured a consistent
application of vidhi rules, but consistency in rule application may be achieved by means other
than the definition of axioms.
Third, it is incorrect to understand the generation of words in the A in terms of recursive
definition, as Joshi seems to be willing to grant:

“Listing the phonemic sequences which I call WFFs (Well-Formed-Formulae) in the Aṣṭādhyāyī
Pāṇini has presented his algebraic formulae on an extremely formal nature, … Pāṇini’s WFFs
can be described as a formal, helpful technique whose application lies in generating the
significant classes of words which he might have anticipated before designing them”20.

As shown earlier, P’s language is not a formal language. It is a highly technical language,
at times, it may even become despairing intricate (think for instance in rule 1.2.28 acaś ca, which
hardly has any resemblance with non technical Sanskrit), but it is still Sanskrit. It is true that due
to the introduction of anubandhaḥ to highlight some morphophonemic changes
(7.2.16,7.2.17,7.3.86), or to form short technical terms (3.3.89), or simply to list of primitive
morphemes as in the pratyāhārasūtras, one might be tempted to see some sort of undeveloped
notion of formal language, but at the end we have to concede that the A performs admirably its
intended task without recurring to a formal language as a metalanguage.

It should be apparent that P gives no recursive definition. The very notion of recursive
definition, mechanical procedure, or algorithm does not belong to the grammatical vocabulary
available to the Pāṇinian grammarians. It is probably accurate, as Joshi and Roodbergen have
acknowledged, to characterize the A’s method (prakriyā) as a “a step-by-step rule-governed
method”21, but from the claim that the prakriyā may be correctly regarded as a step-by-step rule-
governed method it does not follow that the rules of the A might be correctly described as
recursive functions.
19
Joshi 1965/2005, 58.
20
Joshi 1965/2005, 57.
21
Joshi & Roodbergen 1991, 15.

9
 Now if you think of the A as an axiomatic system, you are surely committed to claiming
as well that, being ignored the case for avyutpannaprātipadika, Pāṇini’s “formation of classes is
purely formal and free from semantics”, and so you are inevitably bound to understand the A’s
formation-word rules as recursive functions. Yet, the A itself furnishes evidence enough to reject
that claim. There is no trace as to identify any P’s formation-word rule as a function that calls
itself in the application. It is true that rules 4.1.1 to 4.1.2 have, as Sharma has observed22, a cyclic
domain, and that result of applying those rules may be arguments for other operations, and so
they appear as suitable candidates for being recursive.
Finally, the analogy between the concept of WFF, as used in the language of first-order
logic and the formation word rules in the A is fundamentally flawed. In first-order logic the
WFFs are generated by recursive definition in order to assure that no formula belonging to the
language may be generated by means other than those specified in each one of the definition. In
other words, for any formula A, if A is a WFF, then either A is an atomic formula, or A has been
derived by applying the formation rules defining logical operators.
It appears, then, that the meaning of the WFFs depends solely on the logical relations
stated by each transformational metarule. Unlike the statements of natural language, you do not
need to know the meaning of ‘p’ and ‘q’ in order to understand the meaning expressed by the
statement “((p ↔ q) & ¬p) → ¬q)”. In this case, you may say that the meaning of logical
statements, or theorems in the system, is recursively generated, since the transformational rules
are also, so to speak, meaning-generating rules.
In assuming as WFFs the set of expressions that has been generated according to the
definitions Joshi enumerates in (i)-(v), he is imposing on the grammatical language of the A
syntactical constrains, which are meant to assure that the “formation of classes [of words] is
purely formal and free from semantics”23. Yet those syntactical constrains impose a particular
conception of the A’s grammatical rules, which ultimately result in a severe distortion of its
language.

The etymological analysis of the word “vyākaraṇa” shows that vyākaraṇa is derived
from the root ‘kṛ’ plus the prefixes ‘vi’ and ‘a’ together with the suffix ‘lyuṭ’. In Mahābhāsya,
Patañjali defines “vyākaraṇa” as vyākriyanteśabdā anena (“that by which words are
analyzed”)24. The primary meaning of “vyākaraṇa” seems to be “analysis”, while its secondary
meaning is the “explanation of the words by analysis”. Patañjali’s definition seems to rest on
three assumptions that have not been questioned by the subsequent commentary literature. The
first assumption, the most fundamental one, is that words have already meaning; therefore

22
Sharma 1987,vol.I, 92.
23
Joshi 1965/2005,58.
24
Patañjali 1985 vol.. I: 11.

10
grammatical rules, unlike operational logical rules, are not to be intended as meaning generating
rules, since the meaning is given by its use (loka). That seems to be the underlying
presupposition of rule:

1.2.56 pradhānapratyayārthavacanam arthasyānyapramāṇatvāt [‘a statement to the

effect that the meaning of a suffix is the primary thing (need not be taught) because meaning is
decided by something else (than rules of grammar)’.]

The second is that words may be split into several parts through the process of analysis.
The profusion of technical terms might have been conceived as a formal device to identify the
syntactical elements accounting for the changes that undergo the units out which more complex
structures are built when language is used as a means to communication. A verb, for instance,
when used in normal speech reveals a highly complex syntactical structure, the analysis of which
requires the use of technical terms to single out the types of syntactical changes: the use of l-
markers related to changes of tenses and moods ( laṅ 3.2.111, liṭ 3.2.115, laṭ 3.2.123), or the
assignment of parasmaipada and ātmanepada suffixes (1.3.12 to 1.3..93).
Finally the third assumption is that by stating a set of rules that may eventually include
examples, it is possible to understand the process of word formation in terms of the operations
stated by the rules. These operations, despite of Joshi’s strong claim, are not purely formal and
free from semantics. As Subrahmanyam has noticed:“The meaning of the suffix is taken as
primary and the suffix is assigned for the purpose of expressing the meaning that is conveyed by
it”25. It is not clear if the tiṅantas can be also derived independently from a syntactical context26.

V
There are certainly some undeniable structural similarities between the A and the current strand
of generative grammar prevailing in the late and 1970s, or at least until the Minimalist approach
began to take shape27. And yet, when the A is set against the background of the conception of
language adopted by the Paninian grammarians, both claims (a) and (b) are flawed. They
amount merely to an historical misconception. This misconception rests on the assumption that
cognitive categories belonging to the XX century linguistic paradigm may be employed to
account for the meaning in which P understood the A’s rules, without incurring in a severe
distortion of the text. From claim (a) it follows an additional premise, that, although, not
originally stated as a part of the general argument to prove that the A is a generative grammar in
a Chomskyan sense, may be used to make my point:

25
Subrahmanyam 1999, 41.
26
For Roodbergen’s argument see Roodbergen 2012,21.
27
See: Staal 1966, Joshi 1969 and Kiparsky & Staal 1969.

11
 (a1) “The A is an ingenious device, a yantra, designed to reproduce the language of the
śiṣṭas in a step-by-step rule governed method. In fact, the A may be regarded as an
algorithm, a problem-solving procedure. The problem each time is the derivation of word
ready for use in a sentence. That is why traditionally the A is termed śabdānuśāsana
‘instruction in (the derivation of correct Sanskrit) words’”28.

Before examining he implications between (a) and (a1), few comments need to be made
to clarify some ambiguities. It should be noticed that “ingenious device”, yantra, “algorithm”
and “problem-solving procedure” do not mean the same thing. You may equate algorithm with
problem-solving procedure; but an algorithm is not the same as an “ingenious device” or a
yantra. While it is true to affirm that it is “a step-by-step rule governed method”, it does not
follow that it has to be an algorithm. In fact, the very notion of algorithm is totally foreign in the
linguistic vocabulary of the Pāṇinian tradition. Hardly then the śabdānuśāsana can be correctly
characterized in terms of a notion that was unknown to the Indian grammarians.

So, if the A is to be considered a generative grammar in a Chomskyan sense, then it

would follow also that P’s grammar is also an algorithm, or a problem solving procedure with
the ability to generate a potentially infinite number of sentences in the e-language represented by
Sanskrit. As explained before, the claim that A’s rules are descriptive can be correctly grounded
on the assumption that P makes somehow use of some derivational device that may be
meaningfully compare with the notion of recursive function or algorithm.

The notion of algorithm plays a central role in generative linguistics; essentially it is in

terms of which generative grammar rules have been recognised as descriptive rules. To say the
generative grammar rules are descriptive is to assume that sentences are generated by a recursive
procedure that is considered part of the brain/mind29. This procedure works like command lines
of computer program defining the formal syntactical order for the different parts of a sentence to
fit. Stephen Pinker commenting on the trick of recursion remarks:
28
Joshi & Roodbergen 1991,15.

29
Recursive procedure in this case should be taken to mean Turing Machine (TM). Alan M. Turing
proved in 1936 that there was not an effective method to solve all mathematical problems. The proof is
largely based on the building of an “Universal Turing Machine”, which subsequently was used as an
artificial model for human mind. With it, Turing gave the final blow to the general epistemological
presupposition still prevailing in the first half of the XX century that mental images, numbers and
relations in general could not be represented in the brain unless they were expressed by words. In 1959,
Chomsky proved that the class of Turing-computable functions and the class of grammatically-
computable functions are the same. This result led Chomsky to the solution of the problem that
eventually gave rise to generative grammar: what sort of mental structure is underlying the
concatenation of phrases in the way it actually appears in using language? Chomsky’s insight
was to use the symbolic representation provided by TM as the cognitive structure to account for
language formation and acquisition.
12
 "With a few thousand nouns that can fill the subject slot and a few thousand verbs that can fill
the predicate slot, one already has several million ways to open a sentence. The possible
combinations quickly multiply out to unimaginably large numbers. Indeed, the repertoire of
sentences is theoretically infinite, because the rules of language use a trick called recursion. A
recursive rule allows a phrase to contain an example of itself, as in She thinks that he thinks that
they think that he knows and so on, ad infinitum. And if the number of sentences is infinite, the
number of possible thoughts and intentions is infinite too, because virtually every sentence
expresses a different thought or intention”30.

Generative grammar rules may be then said to be descriptive because they display the
recursive procedure –imbedded in the brain/mind- which is responsible for the human ability of
speaking a language. This ability is called in generative linguistic Universal Grammar (UG),
Human Language Capacity (HLC), or more technically I-Language, and it shows the structure-
dependent character of syntactical rules as the salient feature of the language faculty.

It seems that the HLC poses an insurmountable challenge to the theoretical assumptions
underlying claims (a)-(a1)-(b). In particular, it would follow that the A is based on some sort of
algorithm, that -though not explicitly recognised by P, or the subsequent commentators- aims to
explaining the human mind’s ability of producing a potential infinite number of sentences.
Indeed, the HLC contradicts one of the fundamental tenets of the conception of language held by
the Pāṇinian tradition. Words both vaidik and laukika, exist of themselves and are not created by
anything similar to a human faculty. The relationship between word and object and its meaning
is eternal, and consequently human mind lacks the ability to create language. The point is
illustrated in Mahābhāṣya:

“... How is it known that word, object, and the relationship between the two are eternal? From
the world, for, in the world, one, thinking of various objects, uses words, but they do not make
any effort to coin the words. In the case of objects, however, which are produced, effort is made
to produce them. Thus, for instance he who wants to do something with a pot goes to pottery and
says ‘make a pot, I shall do something with it! But he who wants to use words will not go to a
grammarian’s shop and says ‘make words, I shall use them!31”.

According to Patañjali, words and objects and the relationship holding between them are
siddha, and siddha is “synonymous with “eternal”32. The example of the pottery shows that,
what otherwise may be taken to be a clear example of HLA, -words unlike pots are used by

30
Pinker 2003, 37.
31
Patañjali 1985 vol I. 7‐8
32
Patañjali 1985 vol I., 6.

13
speakers effortlessly- is, nonetheless taken by Patañjali as a reason to claim that words are
eternal. From that claims it follows that human mind plays no role in creating language, at least
in a Chomskyan sense of generating the words and sentences of Sanskrit. On that ground, claims
(a)-(a1)-(b) should be dropped, as they presuppose a particular conception of language that was
not actually held by the Sanskrit grammarians, namely: that HLA, understood as the ability to
coin the words, is not a faculty of human mind.

Finally, there is still a further claim that Joshi and Roodbergen make, that has needs to be
examined, as it seems to contradict both claims (a) and (b). The claim in point is:

(c) “Since it reproduces the standard speech, the A is prescriptive grammar. It states
the rules which must be applied, if the speaker wants to convey the meaning in a
grammatical correct form”33.
The trouble with claim (c) is that it cannot be consistently held with both claims (a) and
(b) at the same time. Much of the discussion about prescriptive and descriptive rules came about
in connection with some specific claims that initially generative grammarians made about the
nature of generative rules. One of those claims was to try to differentiate generative rules from
prescriptive grammatical rules, that were traditionally considered to be prescriptive. Commenting
on the role prescriptive rules play in the language, Pinker has noticed that:

“The legislators of "correct English," in fact, are an informal network of copy-editors,

dictionary usage panelists, style manual writers, English teachers, essayists, and pundits.
Their authority, they claim, comes from their dedication to implementing standards that have
served the language well in the past, especially in the prose of its finest writers, and that
maximize its clarity, logic, consistency, elegance, precision, stability, and expressive range.
William Safire, who writes the weekly column "On Language" for the [New York Times
Magazine], calls himself a "language maven," from the Yiddish word meaning expert, and this
gives us a convenient label for the entire group”34.

Prescriptive grammar rules make statements about how potential speakers should use the
language. We may think of prescriptive rules as if they filter out some part of the entire output of
the language that is socially unacceptable. “Do not split infinitives”, “Use whom not who”, or
“Never end a sentence with infinitives” are some examples of prescriptive rules. It seems that if
P’s rules are prescriptive rules, then they would not describe how the śiṣṭas are using the
language, independently whether they use it grammatically or ungrammatically. In particular,
operational rules would be stating necessary operations a speaker is bound to use to convey its

33
Joshi & Roodbergen 1991,15.
34
Pinker 1995,372‐3.

14
vivakṣa in a grammatical correct form. If so, they would not be able to generate words or
sentences, and that case the A would fail to be regarded as a generative grammar in a Chomskyan
sense. It appears then that claim (c) contradicts both claims (a) and (b). That would make P a sort
of Sanskrit language maven, a picture that would certainly leave much disappointed those who
have seen the A as an insightful anticipation of a generative grammar.
Consider now for the argument’s sake Chomsky’s famous example35:
(6) (a) Colorless green ideas sleep furiously
(b) *Furiously sleep ideas green colorless

Any native English speaker, just in virtue of his ability to understand and speak English, is in a
position to perceive the difference between (6a) and (6b*). Clearly, sentence (6a) expresses an
absurd proposition, and yet it is a grammatically correct English statement, as it conforms to the
generative rules governing the syntax of English, while statement (6b*) contradicts the
syntactical rules for English. Chomsky’s example has been used in generative grammar to
account for the knowledge native English speakers have of their own language and the structural
properties underlying the process of acquiring the language.

The UG aims to specify the actual properties in terms of which the linguistic ability of a
native speaker may be accounted for. Basically, the UG contains:

(i) a set of absolute universals, notions and principles that are supposed to be fixed in
human language, and

(ii) the specific properties of each particular language defined by the empirical range of
each e-language.

Absolute universals are considered to be rigid, they cannot be learnt and are presumed to be in
our genetic code. That seems to be one of the reasons to talk about the biology of human
language or biolinguistics. According to general principles of UG, language –strictly speaking- is
not learnt, it is rather acquired. The knowledge of a native speaker has of his own e-language is
essentially an unconscious knowledge and it assumed to be somehow wired in the brain/mind . It
follows that no native speaker is aware that he is actually using a rule system governing the
syntactical properties of phrase-construction and consequently he is not expected to have direct
access to that sort of knowledge by introspection. On this account, as shown above, generative
grammar rules said to be descriptive: they describe the innate/unconscious knowledge that
native speakers display in using their e-language, as opposed to prescriptive rules that are based,
to put it in Pinker’s terms, on standards that have served the language well in the past. These
standards do not belong to the linguistic genetic endowment, and the rules they prescribe are

35
Chomsky 1957, 15.

15
intended to preserve the particular pattern of linguistic behaviour as exhibited in a speaking
community.

Similar cases to (6a) may be found in Sanskrit as well. Nāgeśa uses the following
example:

(7) eṣa bandhyāsuto yāti khapuṣpakṛtaśekharaḥ

kūrmakṣīracaye snāta śāśasṛṅgadhanurdharaḥ

[There goes the son of a barren woman with his hair-top bedecked with sky-flower,
bathed with milk of a tortoise carrying a bow made of the horn of a rabbit]36.

If P’s rules are descriptive, then Sanskrit statements similar to (6a) would be generated by the
grammar. According to Sharma, however, statements like (7) are:

“grammatically approved though treated as kalpanā ‘imagination’ and hence, regarded as falling
outside the scope of normal usage. Since Sanskrit grammarians do not bother themselves with
the intuitive knowledge of the native speakers, they do not consider it necessary to go deeper into
grammatical semantic deviance. They adhere to the usage of the śiṣṭas as norm”37.

As noted earlier, no conclusive evidence is available to support if P’s grammatical rules

were considered by the Indian grammarians descriptive, o prescriptive. Probably the existing
evidence provided by the commentary tradition may be used to support one view or another. I
suppose you could say that problems related with deviant uses, or corrupted words (apaśabda)
might have been resolved by providing a well-defined set of grammatical rules according to
which doubts about the correct syntactical form of words could have been settled. In that case,
P’s grammatical rules may be considered as prescriptive rules.
If everyone in a speaking community consistently uses words and sentences in a way that
is grammatically acceptable, then there would be no need for explicit prescriptive grammatical
rules to ensure that the grammar of the language is not contravened by deviant uses. Now, if at
some point there was perceived that the use of mleccha language by some brahmins and the
increasing presence of corrupt words (apaśabda) were compromising the purity of Sanskrit38,
then it would be reasonable to assume that prescriptive rules come to be formulated in order to
keep the unacceptable linguistic forms in check.
Under this view, Pataṅjali’s claim about the purpose of the grammar evam eṣā
śiṣṭajṅānārthāṣṭādhyāyī might be interpreted as endorsing a prescriptive meaning for

36
Quoted in Sharma 1987, 50.
37
Sharma 1987 ,50.
38
Joshi & Roodbergen 1991,8.

16
grammatical rules, in terms of which the authoritative source of Sanskrit spoken by the śiṣtas is
codified by the A. Grammatical rules then, may be viewed as prescribing the necessary
operations on the selected grammatical items, in order to ensure that the speaker’s intention
conforms with the established linguistic behaviour exhibited by the śiṣṭa community.
If during P’s life-time linguistic variations were common in the spoken Sanskrit, a
reasonable guess would be to expect the formulation of prescriptive grammatical rules. It is then
very likely that the A could have been conceived as a prescriptive grammar, and as such P could
not have treated all the occurring variants as equally acceptable. The descriptive or prescriptive
meaning of the A’s rules would be then determined, as Kiparsky has argued, by the way in which
vā and vibhāṣā are interpreted: “if the grammar is prescriptive, these terms can mean
“preferably” and “marginally”. If it is purely descriptive, then only “frequently” and “rarely”39.
From the argument above given, however, it does not follow that the A is not a
descriptive grammar. What it does follow is that P’s rules cannot be meaningfully regarded as
descriptive rules, when the A is assumed to be a generative grammar in a Chomskyan sense. On
the whole, claims (a)-(a1)-(b) force a particular interpretation on the A that:
(i) does not seem to be supported by the general conception of language developed by
Sanskrit grammarians,
(ii) is based on the mistaken historical assumption that the meaning of ancient texts can
be correctly understood in terms of a vocabulary that was totally foreign to the conception under
which those texts were written.

REFERENCES

Cardona,G.(1980). Pāṇini: a survey of research. New Delhi: Motilal Banarsidass..

Chomsky, N. (1957). Syntactic structures. The Hague/Paris: Mouton.

__________ (1965). Aspects of the theory of syntax. Cambridge, MA. The M.I.T Press.

Coward,H.G. & K.Kunnjunni Raja,.K.K.(1990). Encyclopedia of Indian Philosophies, vol. V:

The Philosophy of the Grammarians. New Delhi: Motilal Banarsidass.

Joshi, S. D. (1965/2005). Two methods of interpreting Pāṇini. Poona: Publications of the Centre
of Advanced Study in Sanskrit, University of Poona.

_________ (1969). Sentence structure according to Pāṇini. Indian Antiquary III.3: 14-26

39
Kiparsky 2012, 327.

17
Joshi, S. D. & Roodbergen, J. A F. (1991). The Aṣṭādhyāyī of Pāṇini with translation and
explanatory notes, vol. I (1.1.1 – 1.175). New Delhi: Sahitya Akademi.

Kiparsky, P.& Staal, J.F.(1969). Syntactic and semantic relations in Pāṇini. Foundations of
Language 5: 83-117.

Kiparsky, P.(2012). Pāṇini, variation and orthoepic diaskeuasis. AS/EA LXVI: 327-335.

Pinker, S. (1995). The language instinct. New York: Penguin Books.

_______ (2003). The Blank Slate. New York: Penguin Books.

Patañjali (1985). The Vyākaraṇa-Mahābhāsya of Patañjali, vol.I. Edited by L.F.Kielhorn,Poona:

Bhandarkar Oriental Research Insitute.

Roodbergen, J. A .F. (2012). The Aṣṭādhyāyī of Pāṇini: A brief exposition. New Delhi: Adarsha
Sanskrit Shodha Samastha.

Scharfe, H. (1971). Pāṇini’s Metalenguage. Memoirs of the American Philosophical

Society,No.89. Philadelphia: American Philosophical Society.

Sharma,R.N.(1987).The Aṣṭādhyāyī of Pāṇini,vol.I. New Delhi: Munshiram Manoharlal.

Staal J. F. (1961). The theory of definition in Indian logic. Journal of the American Oriental
Society, Vol. 81, No. 2: 122-126

_____ (1965) Euclid and Pāṇini. Philosophy East and West 15,No.2: 99-116.

_________(1966). Room at the top in Sanskrit: ancient and modern descriptions of nominal
composition. Indo-Iranian Journal 9: 165-198.

Subrahmanyam,P.S.(1999).Pāṇinian linguistics.Tokyo: Institute for the Study of Languages and

Cultures of Asia and Africa.

18