© SHESL/EDP Sciences

1 Preliminaries

This paper will deal with some aspects of Indian grammars elaborated between the 7^th and the 17^th c. to edict rules or describe some of the linguistic stages that come under the broader heading ‘Middle Indo-Aryan’ (or Middle Indian). This designation refers to the intermediate linguistic stage between Old Indo-Aryan, namely Vedic and classical Sanskrit, and New Indo-Aryan which culminates ultimately in languages such as Hindi, Bengali, Marathi, Gujarati, etc. But one should not picture the relation between Old Indo-Aryan and Middle Indo-Aryan as chronologically layered.¹ They are distinct stages, but from the earliest times Old Indo-Aryan has been influenced by Middle Indo-Aryan and there are close linguistic connections between both. It is admitted that at the same time a more polished speech (Sanskrit) was used in particular contexts and situations and coexisted with more vernacular dialects (Middle Indo-Aryan). Sanskrit, for which the absolute norm became Pāṇini's grammar composed not later than the 4^th c. BCE, appears as the language of the intellectual elite, represented by the Brahmins. Middle Indo-Aryan, on the other hand, refers to the languages that were spoken by larger sections of the population. They spread even more from 500 BCE because they were promoted by the Buddha, and by Mahāvīra, the founder of Jainism. The oldest evidence of Middle Indo-Aryan dates back to the 3^rd c. BCE with the rock and pillar inscriptions engraved at strategic places of the Ashokan empire. The grammars on which this paper focuses deal with literary languages of Middle Indo-Aryan:

• Pali is the language of the Buddhist scriptures belonging to the Theravāda school (1^st c. BCE + in written form) and represents the oldest linguistic stage of Middle Indo-Aryan.

• The term Prakrit refers to a plurality of related languages or dialects that have specific names and are used in various literary contexts: in classical Indian plays, in poetry, in stories or in the religious literature of Jainism. So it is better to speak of the Prakrits. The grammars of the Prakrits generally take one of them, called Māhārāṣṭrī, the language of poetry par excellence, as the standard for description and devote separate sections to the special features of the other Prakrits. Here only facts relating to the treatment of this standard Prakrit will be discussed.

First, the most common types of linguistic differences between Sanskrit and the various Middle Indian languages are summarized here:

Phonetic features:

• Vocalic r is replaced by a, i or u.

• Diphthongs ai and au are replaced by e and o.

• Long vowels before two or more consonants are shortened.

• Three sibilants ś ṣ s are reduced to one, either ś (Māgadhī Pkt.) or s (elsewhere).

• Complex consonant clusters are assimilated into geminate consonants or split through epenthetic vowel, e.g. Skt. sarpa ‘snake’ vs MIA sappa; Skt. ratna ‘jewel’ vs MIA ratana.

• Single intervocalic stops are weakened or removed, e.g. 3^rd pres. singular ending -ati (Pali), -adi (Śauraseni Pkt.), -ai (elsewhere).

• Dentals are palatalized by following -y-, e.g. Skt. satya ‘truth’ vs MIA sacca.²

Structural features:

• Case endings and developed inflections are at the same time same and different; there is a profusion and accumulation of forms existing side by side (older ones and new ones) but case endings differentiations are reduced in the latest stage of MIA (Apabhramsha).

• Three numbers: singular, dual, plural > MIA two numbers; dual is no longer alive and thus represents an irrelevant category.

• Developed verbal system > same and different; profusion and accumulation of forms existing side by side (older ones and new ones).

• Variety of past tenses > progressive disappearance of several conjugated past tenses.

• Identical standard word order S O V in Sanskrit and MIA.

• Identical use of gerund to expand sentences in Sanskrit and MIA.

• Identical use of two balanced clauses as main subordination system in Sanskrit and MIA.

• Identical absence of indirect speech in Sanskrit and MIA.

The purpose here is only to highlight roughly the type of linguistic relations there are between all these languages, and the overall closeness they exhibit. In short, they are distinct but not fully alien to each other in their forms or grammatical structures. This point is meaningful if we turn to the issue of the metalanguages used by the grammars of Middle Indo-Aryan.

Grammatical corpora of MIA

Only the main landmarks of the vast corpus that grammatical works of MIA represent are listed here, underlining the striking difference that needs to detain us here between Pali and Prakrit grammars.

In all grammars of Pali the language of description is Pali:

• Kaccāyana's grammar (7^th-10^th c.), spread in all Theravāda countries.³

• Saddanīti by Aggavaṃsa (12^th c.), composed in Burma.⁴

• Moggallāna's grammar (12^th c.): represents a different trend, spread in Sri Lanka.⁵

In all grammars of Prakrits the language of description is Sanskrit:

• Vararuci, the oldest Prakrit grammar (3^rd-4^th c. CE), very popular in various regions of India.

• Eastern school of grammarians represented by Puruṣottama (end of 11^th-beginning of 12^th c.), Mārkaṇḍeya (17^th c.), Rāmaśarman (17^th c.).

• Western school represented by two Jain authors: Hemacandra (12^th c.) in Gujarat and Trivikrama (second half of the 13^th c.) in South India (Andhra Pradesh).⁶

All grammars conform to the traditional sūtra form, namely concise aphorisms. Only two of them (Trivikrama and Rāmaśarman) put the contents into verses, which means adding verbs like ‘are’, ‘should be’ instead of keeping only to purely nominal sentences. All these grammars take recourse to the metalinguistic use of cases, to termini technici (saṃjñā), and to techniques such as immediate recurrence (anuvṛtti) of a word from one aphorism to the next, all features met with in Sanskrit grammatical works, since Pāṇini's Aṣṭādhyāyī, the model par excellence, but not the only one.

2 The choice of Metalanguage in Grammars of Middle Indian Languages

Pali is described through Pali so that all Pali grammars are written in Pali. But Prakrits are described through the medium of Sanskrit. Thus the grammars of Prakrits can be described as ‘extended Sanskrit grammars’ in the strict sense: they use a language that has acquired an impressive cultural aura especially in scholarly areas in order to describe languages that are different from it. But the grammars of Pali come under this heading too: although they use Pali and not Sanskrit for the description, the descriptive model is clearly that of Sanskrit.⁷ Given the type of relationship between Sanskrit and Middle Indo-Aryan, the situation of extended Sanskrit grammars of Middle Indo-Aryan is quite different from extended Sanskrit grammars applied to Indian languages of other families, such as the Dravidian languages, where there is a wide structural gap between the language of description and the language described.

What could be the reasons for the contrastive situation in metalanguages we observe between Pali and Prakrit grammars? They largely relate to the perception the grammarians of these traditions have of the languages they describe. This presides over their rejection or adoption of Sanskrit as their metalanguage.

2.1 Perceptions of Pali

I select Aggavaṃsa (the 12^th century Burmese monk author of the Saddanīti) as my representative of the Pali attitude, for his grammar has the advantage of providing us with a developed and coherent discourse in his own commentary of the sūtras. Right at the start he explains that his work is meant to elucidate all the implications of the Buddha's teaching (buddhavacana) so that there is no confusion on letters and words. As a Theravāda Buddhist, his belief is that the Buddha preached in the language Buddhists call Māgadhī, that is Pali. His grammar is imbibed with an extremely large number of quotations from the scriptures with which he had an unmatched familiarity. So, for him, the question even does not arise of using a language other than Pali. The same holds true for his predecessor, Kaccāyana, but the latter seems to take it for granted and does not elaborate on this point.

In addition, Aggavaṃsa also announces his desire to cleanse the language (visodhana), as he says, of forms that he considers foreign to the Buddha's speech.⁸ They are those coming from the pressure of Sanskrit against which he raises his voice. He cannot fight against admitted Sanskrit loanwords that have crept in Pali and look unusual because they contain a consonant cluster that has not gone through assimilation. Among them there are words that are used only in their Sanskrit form such as indriya ‘sense organ’, brāhmaṇa, ‘brahmin’, as well as some which can be used either in their Pali form (with consonant group disjunction) or in their Sanskrit form (with consonant group maintained) such as ratana versus ratna ‘jewel’, paduma versus padma ‘lotus’, kilesa versus klesa ‘defilement’.⁹ The grammarian can only note the existence of those belonging to the first group, as there is really no other choice: such words are found all over in the texts reflecting the Buddha's teaching. For words known under two forms, however, he tries to delineate a distribution principle, noting that the Sanskritic forms are mostly found in verses.¹⁰ Similarly, he separates himself from others who recognize the Sanskritic form guru in the meaning ‘teacher’ and the Pali form garu in the meaning ‘elders’, only admitting garu and using it throughout his grammar.¹¹ He criticizes his predecessor Kaccāyana who postulates forms such as uccate ‘is said’, present passive 3^rd singular, and uttaṃ ‘said’, past participle, and opposes to them the expected Pali forms, respectively vuccati and vuttaṃ:¹² the former are mechanical transpositions of Skt. ucyate and uktam, whereas the latter exhibit Pali phoneticism with initial epenthesis and a Pali morphological feature for vuccati, with the active ending -ti used for the passive as it is common in Pali. But there is a certain amount of fluctuation: in one place he considers the Sanskritic form kriyā ‘action’ as a borrowing and kiriyā as the only valid form in Pali, but, elsewhere, however reluctantly, he admits kriyā relying on the authority of the ‘commentaries-authors’.¹³ His general stand, which is to criticize the scholars deceived by their familiarity with Sanskrit books, leads him to reject Sanskritisms in individual words¹⁴ and to state specific rules, for example in the area of sandhi¹⁵ where the pressure of Sanskrit was probably high. All this shows how uneasy the fight to defend the Pali linguistic identity is and how the Sanskrit influence acts as a diktat.

The Sanskrit grammarians and, to some extent, the later Pali commentators, are Aggavaṃsa's favourite targets. This can be seen, for instance, in his discussion of the dual.¹⁶ This grammatical number is no longer alive in Middle Indo-Aryan, and thus represents an irrelevant category in this linguistic stratum. But the occurrence of a specific form of the pronoun used in connection with two monks in the Vinaya (books of monastic discipline), contrasting with another form in the case of three or four monks, raises a question. Aggavaṃsa violently goes against those who would claim on the basis of this form that the dual exists in the Buddha's speech. This form is specific of the Vinaya language, he says, but it cannot be said to represent a general situation. His conclusion is clear: ‘The dual number exists in the Sanskrit language for two (items), but in Pali it does not exist’.¹⁷ This leads to what seems to be the coinage of a new grammatical term, i.e. puthu-vacana as a designation of plural (instead of the common Sanskrit term bahu-vacana also used by other Pali grammarians). This word is supposed to convey at once ‘the difference’ there is in what plural means in Pali and in Sanskrit.¹⁸ In Sanskrit, dual is used for 2 not for 3, 4, 5, etc., but in Pali the plural is used for 2, 3, 4, 5, etc.

The discussion of grammatical gender is another of his concerns. For instance, he establishes the masculine gender and the singular number of the word āpo ‘waters’ on the basis of quotations. But in one passage of a late sub-commentary (ṭīkā) this word is found used in the feminine plural¹⁹ – like in Sanskrit. Although exonerating the sub-commentators and paying respect to them (116,3-5) the grammarian says:

These (subcommentators) were not only conversant with the Buddha's speech, the Tipiṭaka and its commentaries, but also with the whole of (Sanskrit) grammar. That is why, having borrowed the rule found in Sanskrit grammar, they have described the word āpa as a feminine plural in order to show their own proficiency in grammar, and to make known to others that such was the rule in Sanskrit grammar'.²⁰

One of the reasons he finds to this situation is that there is ‘sometimes a complete disagreement between the rule of the sacred text and the rule of the Sanskrit grammarians’,²¹ with the result that ‘in Sanskrit grammar the word ‘water’ is feminine plural, in the speech declared by the Buddha it is masculine singular’.²²

So he obviously sees the pressure of Sanskrit, increasing in his time, as a threat and behaves as a Pali purist. To sum up: Pali grammars are meant to elucidate all the implications of the Buddha's teaching. For Theravāda Buddhists, the language of the Buddha is what they call ‘Māgadhī’, what we call Pali. Therefore, no other language needs consideration. The grammars contain numerous quotations from the scriptures and are viewed as instances of the Buddha's word. Sanskrit and ‘Sanskrit grammars’ (Pa. sadda-sattha / Skt. śabda-śāstra) are felt as a pressurizing threat, especially in Middle Ages South East Asia where Sanskrit language and its literary tradition expand their influence.

2.2 Perceptions of the Prakrits

Let us now come to Prakrit. The first work where Prakrit is somehow dealt with is the Nāṭyaśāstra – not a grammar, thus, but a treatise on poetics and dramaturgy. In brief, in classical Indian plays, Sanskrit is spoken by some characters and Prakrit by others. In this context, Prakrit is considered ‘as the spoken form of Sanskrit, with deviating pronounciation here and there. Hence Prakrit is in Bharata's opinion not a different language; Sanskrit and Prakrit are different variants of the same language'.²³

Several grammars in the corpus start in medias res with the treatment of vowels which invariably is their first topic.²⁴ They have no preamble of any sort or any thought on the process of writing etc. In such cases, the fact that their starting point is Sanskrit, however, becomes clear from statements found here and there such as: ‘The roots which are known here (= Sanskrit) as ending in consonants will be changed in Prakrit to roots ending in a’, says Rāmaśarman, for instance.²⁵ This is a way to express the process of transfer to thematic paradigms which is a prevalent morphological feature in Prakrit and one of the main differences with Sanskrit.

Other grammars, on the other hand, offer as a preliminary an interesting discourse which throws light on their perception of Prakrit. They clearly view it as ‘derived’ from Sanskrit. They do recognize the existence of words that are not connected with Sanskrit, but they do not consider them as the subject of their grammars. Authors who treat these words do it in special works, such as dictionaries of regional words.²⁶ I here briefly review a few famous passages that have already been dealt with in scholarship.

Hemacandra's grammar of the Prakrits comes after seven chapters devoted to Sanskrit. Hence he starts ‘And now Prakrit’ and explains:

The basis is Sanskrit. That which is located there or has come from there is Prakrit. Immediately after Sanskrit Prakrit is established as the new topic. And the fact that teaching about Prakrit comes immediately after Sanskrit has the purpose of implying that the body of rules is only for that which has Sanskrit as its origin in either of its (two) categories: words that have already been formed²⁷ and those that are in the course of being formed,²⁸ not to regional (words). As for what is identical with Sanskrit, it has already been explained through the rules meant for Sanskrit. In Prakrit too one speaks of base, affix, gender, cases, compounds, technical terms, etc.²⁹

The grammarian then lists obvious and crucial general differences between Prakrit and Sanskrit in terms of phonetics and morphology: the sound system is the same except that vocalic ṛ, short and long, vocalic l, short and long, the diphtongs ai and au, the guttural and palatal nasal, the palatal and retroflex sibilant, the visarga and the plūta do not exist in Prakrit. The guttural and palatal nasal are found only in consonant groups; ai and au may be found occasionally. A consonant without vowel is not found. The dual and the dative plural are not attested.³⁰ Thus we see right at the beginning that Sanskrit is viewed as the explicit and obvious model and that the purpose of the grammar is to consider the features that make Prakrit different from Sanskrit.

Trivikrama, who works along the line of Hemacandra, has a similar discourse in the opening verses of his grammar and in the first sūtra. But he first states that a word which has much meaning and which is easily pronounced is the soul of poetry and it is also Prakrit, thereby ascribing a positive value to this language and referring to one of its main applications. He then recognizes that there are forms of the language that depend on use (rūḍha), that they are quite independent and hence do not stand in need of grammar. Only tradition can give information about them. The Prakrit forms and words which are treated in his grammar are those which can be subjected to rules of grammar and can be traced to fully formed Sanskrit words (siddha) or words in the course of being formed (sādhyamāna).³¹

Mārkaṇḍeya in the 17^th c. is in tune with this position, which he repeats at the outset:

Sanskrit is the base. What comes from there is called Prakrit. There are two kinds of words: originating from it (i.e. Sanskrit) and identical to it, and some say there is also regional (vocabulary).

Commentary: A tadbhava exists in Sanskrit and has assumed another form through rules, like rukkha ‘tree’, ghara ‘house’, peranta ‘end’. Examples of a tatsama are tarala ‘trembling’, taraṅga ‘wave’, manthara ‘slow’, etc. A regional (word) is not realized by the rules, it is known in this or that region and used by great poets, such as laḍaha ‘handsome’, peṭṭa ‘belly’ or tokkha ‘?’.'³²

This is a tripartite classification which dates back to the 7^th c.³³ and has become a common tool to describe not only Prakrit but also New Indo-Aryan vocabulary.

One caveat about such passages: the grammarians take into consideration the variety of Prakrit and especially of its vocabulary which draws upon all sorts of sources. What they mean to say is that the focus of their grammars is on those words of the language for which rules can be edicted and systematized. This is why they concentrate on substitution rules. This probably reflects their aim, namely to provide the knowers of Sanskrit with the equipment allowing them to understand how Prakrit works, and this motivation also accounts for the choice of Sanskrit as metalanguage. Whether these grammars alone are sufficient to enable anybody to write fluent poetry of high quality in Prakrit is however doubtful. The overwhelming number of substitution rules from Sanskrit to Prakrit characterizing these grammars is probably what has contributed to spread a kind of mechanical vision of Prakrit as restricted to a language having a derivational status.

Nevertheless, other challenging views on what Prakrit is do exist. An 8^th c. poet, for instance, who wrote an epic poem in Prakrit, is radically different:

All forms of speech enter (Prakrit) and all forms of speech start from it: waters enter nowhere but into the sea and start from nowhere else than from the sea.³⁴

This statement may have to be assessed cum grano salis as the author is a connoisseur of Prakrit who is keen on promoting the language as a fullfledged literary language in no way inferior to Sanskrit. But it is echoed in the 11^th c. by another intellectual, the Jain commentator Namisādhu:

prakṛti is the natural use of speech made by all beings of the world which is not refined by grammar etc.; that which is derived from this prakṛti itself is prākṛta.³⁵

Another internal etymology he provides is prākṛta as ‘that which has been created first’ (prāk kṛta). He then explains how Prakrit and Sanskrit are related:

Prākṛta is the speech easily intelligible to children, women etc. and the source of all other speeches (sakala-bhāṣā-nibandhana-bhūtaṃ vacanam). It is of a homogeneous character like rain falling from the clouds. That same speech, being differentiated according to differences in place and on account of refinement, obtains the subsequent division into saṃskṛta and other speeches. Saṃskṛta is called so on account of being refined by Pāṇini and others, in the rules of their grammar.³⁶

In this approach, Prakrit comes first and Sanskrit second. The prākṛta words are original, and the saṃskṛtas are those which have been transformed. In short, one has to distinguish between a general view on the relationship between Prakrit and Sanskrit and a particular application of it in the context of grammatical rules. But that there is a direct relationship and that one cannot go without the other is a recurring idea, and this is in sharp contrast with the Pali attitude.

3 Grammars of Middle Indian on the verb: how to negotiate with Sanskrit models

Through a case-study, a few aspects of verb-description, we will try to investigate how the grammarians of Pali and Prakrit negotiate between the Sanskrit models and the reality of the languages they describe. How much impact does the selected metalanguage (Pali/Sanskrit) really have?

3.1 Active and middle voice endings

In Sanskrit, each tense or mode is conjugated both in the active voice (parasmaipada) and in the middle voice (ātmanepada) quite systematically. But in Prakrit the middle voice is regressing. Here is how Vararuci deals with the conjugation of the present singular.³⁷

For the affixes of the third person singular in the active and middle voice -i and -e are the endings in Prakrit. Thus paḍhai (active), paḍhae (middle) ‘he recites’ or sahai (active), sahae (middle) ‘he endures’ in front of Sanskrit paṭhati (active), paṭhate (middle) or sahati (active), sahate (middle).

For the affixes of the second person singular in the active and middle voice -si and -se are the substitutes. Thus paḍhasi, paḍhase ‘you recite’ or sahasi, sahase ‘you endure’ in front of Sanskrit paṭhasi, paṭhase or sahasi, sahase. The forms are identical in Prakrit and Sanskrit, except for the voicing of the intervocalic consonant in Prakrit.

Now, for the affixes of the first person singular in the active and in the middle voice, both, -mi is the only substitute taught. Thus Prakrit has only paḍhāmi ‘I recite’, hasāmi ‘I laugh’, sahāmi ‘I endure’ – where Sanskrit would have paṭhāmi, hasāmi, sahāmi in the active and paṭhe, hase, sahe in the middle voice.

Here we see that, although the grammarian of Prakrit undertakes to achieve a systematic presentation where the forms of both the active and middle voice are considered for the three persons and are distinct from each other, he has to give up the binary opposition in the case of the first person singular probably because he takes into account the real usage of the language where this form in the middle voice is not attested. The position of Vararuci, just described, is shared by Hemacandra and Mārkaṇḍeya. ³⁸ But a more radical stand is adopted by Puruṣottama. He prescribes only one substitute for both voices, not only for the first person singular, but also for the second and the third:

Instead of the affixes of the active and middle voice in the third person singular there is -i (6.2). Instead of the affixes of the active and middle voice in the second person singular there is -si (6.4). Instead of the affixes in the active and middle voice in the first person singular there is -mi' (6.7).

This amounts to denying any specific existence to an independent conjugation in the middle voice. Similarly Vararuci does not teach -nte, the third plural ending in the middle voice, which, indeed, is unknown in the dialect he describes.³⁹ Such an example illustrates the tension between the Sanskrit model and pattern of conjugation to which the grammarians of Prakrit are indebted, on the one hand, and the relevance of this model in respect with specificities of the Prakrit that do not fit in, on the other. The solutions are either the compromise or the innovation.

3.2 A special syntactic use of the Prakrit present participle

Given their basic postulate and their perception of Prakrit as being derived from Sanskrit, the grammarians of Prakrit focus predominantly on the description of phonetic and morphologic differences. Syntactic facts are rarely considered.⁴⁰ One instance, however, of a specific syntactic use relating to a given verbal form, namely the present participle, is dealt with by three Prakrit grammarians:

• Puruṣottama 6.32 lṛṅaḥ śatṛvat ‘The conditional is replaced by the present participle’.

• Hemacandra 3.180 nta-māṇau ‘The active present participle and the middle present participle’ (are substitutes for the conditional).

• Mārkaṇḍeya 6.18 lrṅṅarthe vā (śatṛ-śānayor nta-māṇau) ‘The present participle, both active and middle, may also stand in the meaning of a conditional’.⁴¹

Indeed, even if the middle voice is mostly not alive and productive in Prakrit, the middle present participle is common and, as a nominal form, stands apart from the general system.

Hemacandra provides a poetical verse as illustration:

O moon (litt. the one with the deer as a mark), if you had placed the lion (litt. the master over the deer) [on your body] instead of the deer, you would not have had to endure the insult of Rāhu (= the eclipse), the one who wins.⁴²

As for Mārkaṇḍeya, his examples are simpler: ‘If he would read, he would be learned’ or ‘if he would eat, he would feel satiated’.⁴³ This sentence pattern, where the present participle functions as the main verb in both parts of the conditional system, is met with very often in Prakrit and corresponds to a linguistic reality. Although it is occasionally attested in Sanskrit too, it remains marginal and is characteristic of a non-formal style. In the evolution of Indo-Aryan, on the contrary, it tends to become the prevalent pattern.⁴⁴ Such a case shows how using Sanskrit as the metalanguage and conventional grammatical terminology does not prevent the grammarians from taking into account, at least occasionally, facts that are typical of the language they describe.

3.3 Middle Indian and the verbal roots

In the Sanskrit grammatical tradition the verb is based on a root, dhātu. These roots are classified into ten groups (gaṇa) depending on the type of present formation they use.

Let us see the status of these groups in Middle Indo-Aryan grammars. At least two of the most important grammarians of Pali belonging to the same line of teaching, namely Kaccāyana and Aggavaṃsa, make use of this classification, but they contain adjustments to the linguistic reality of the language they describe.

Both of them record only eight groups and not ten: group 2 (root present stems with variation between strong and weak grade) and group 3 (reduplicating presents with variation between strong and weak grade) are irrelevant in the language they describe and are thus left out as groups.⁴⁵ In addition, the sequence of the groups is different from the traditional one. The thematic conjugations, which correspond to class 1, 4, 6 and 10 in the Sanskrit descriptions, are reduced to three groups, because the distinction between thematic presents with and without vowel gradation is abolished. This results into the disappearance of class 6 (tudati) as an independent entity and into its merging into class 1. But, on the other hand, a new class is added, the group gahādi. The root gah means ‘to seize, to take hold of’, corresponding to Skt. GṚH, gṛhṇāti which is a member of class 9. The existence of a specific sūtra for this verb and a wording comparable to other aphorisms describing verb classes make it clear that Kaccāyana considers gaha as the leader of a separate class: ‘After gaha etc. (the suffixes are) ppa and ṇhā’,⁴⁶ hence the present forms gheppati and gaṇhāti. While Kaccāyana remains content with this statement, Aggavaṃsa marks his awareness of the fact that this is not something accepted by all and explains why the existence of a distinct class is justified, referring to Kaccāyana's authority:

Here there are some people who think as follows: indeed, a separate group for gaha etc. is not attested. How could there be a group for gaha etc. with only one root? Because ppa and ṇhā are not found after any other root than gaha, there are some who have included this root in the group of kiy ‘to buy’ (Skt. krī-) without speaking of a group ‘gaha etc.’ A wise man, an expert would not accept their view because Kaccāyana in his rule has established a separate group for gaha etc.⁴⁷

Since the case of this root has not been included in the rule relating to the root ‘to buy’ and since the wording has ‘gaha etc.’ and not ‘gaha’ only, it means it refers to a group. And, he argues, there is, indeed, a group. He then treats verbal forms showing the consonant sequence -ṇha-, such as uṇhati ‘is warm’, uṇhāpeti ‘makes warm’, etc. Whether the demonstration is fully convincing is another matter. But we see the polemic tone in the discussion and the role played by the authoritative reference of Kaccāyana. Making the root gaha the leader of an independent class attracts attention to its specific forms in Pali usage which cannot really be explained by reference to other verb classes.

Finally, the inclusion of a given form within one class or the other may become a matter of discussion in case there is a conflict between the teaching of the (Sanskrit) grammarians and the Pali scriptures. This is what happens with Pa. paccati used as non-transitive in the sentence ‘Devadatta is roasted in hells’.⁴⁸ The meaning requires, Aggavaṃsa says

that the verb belongs to the class div-ādi (intransitive or stative verbs using the suffix -ya-in Sanskrit). If it is objected that the root pac- has not been listed in this group in the (Sanskrit) grammars, the answer is: whether it is or not, how does the (Sanskrit) grammars matter? The sacred text only is the authority. Therefore, considering only the sacred text of the Buddha, who is an expert in idiomatic language, I will put this root also in the group div-ādi.⁴⁹

Since Pāṇini onwards, there are special collections, called Dhātupāṭhas, where the Sanskrit roots are listed according to the above mentioned classification  (Palsule 1955). Such collections do exist also for Pali: the ‘garland of roots’ (Dhātumālā) structures Aggavaṃsa's whole work and is at the centre of five long sections.⁵⁰ In addition at least two separate collections are also known for Pali: a Dhātupāṭha and a Dhātumañjusā.⁵¹ This is in tune with the fact that Pali grammarians acknowledge the tenfold classification as an organizing principle for the Pali verb, even if they adjust it to the requirements of the language they describe.

By contrast, the situation is totally different for Prakrit: we do not have collections of roots. In her seminal monograph Nitti-Dolci tried to explain this absence: first, she argues, Prakrit grammarians must have thought that the students could consult the Sanskrit Dhātupāṭhas and somehow manage to build the Prakrit verbal forms thanks to the analogy with the noun system. In fact, she adds, the Prakrit grammarians were well aware of the difficulty to adapt their often artificial and ad hoc system of phonetic correspondances to varying units like verbal stems. ‘From a single Sanskrit root, she concludes, are derived different themes, that, in their turn are combined with varied affixes: every case falls under the stroke of a different phonetic rule.’⁵² In fact, although the grammarians of Prakrit use Sanskrit as their language of description, they do not seem to be as much influenced by the Sanskrit norm as one could first think when it comes to the verb. In Prakrit cards are shuffled and we see several changes or transfers of conjugational patterns. Here is an example of the way grammarians deal with this issue:

Vararuci 8.29 ciñaś ciṇaḥ: ‘Instead of ci- there is ciṇa’, so present 3^rd person singular ciṇai. The starting point is the root in the form adopted in Sanskrit collections: ciñ cayane.⁵³ Although it is listed in the tenth group, where the present marker is -aya-, hence cayati, this present formation is limited to Vedic Sanskrit, the oldest stratum of the language, while the common present is cinoti, so class 5 according to Sanskrit classifications.

Vararuci 8.56 śru-hu-ji-lū-dhuvāṃ ṇo 'ntye hrasvaḥ: ‘At the end of roots śru “to hear”, hu “to offer”, ji “to conquer”, lū “to cut”, dhū “to shake” ṇa (is to be employed) and (the preceding vowel if long) is shortened, e.g. suṇai, huṇai, jiṇai, luṇai, dhuṇai.⁵⁴

Hemacandra 4.241 ci-ji-śru-hu-stu-lū-pū-dhūgāṃ ṇo hrasvaś ca⁵⁵ adds two more instances to the preceding list. The commentary also expands by adducing the sūtra tool bahulam ‘generally’ (Hc 1.2) in order to cover the doublet forms optionally attested in Prakrit, for example uccei beside ucciṇai or jayai beside jiṇai.

What we see here is that the use of Sanskrit as the metalanguage is no hindrance to a purely synchronic description of verbal forms. Those which behave in the same way in Prakrit are listed together. The fact that they belonged to other conjugational patterns in Sanskrit, which, in addition, are not the same for all the roots listed, is irrelevant at the sūtra level. As for the commentators, they are generally satisfied with listing the present form of the roots, although they occasionally comply with the Sanskrit diktat and give the equivalent Sanskrit form. ⁵⁶ All the verbs of this list illustrate the overwhelming tendency of Prakrit to develop thematic verbal stems, where Sanskrit had roots or suffixes varying between strong and weak grade. The adjunction of the affix -ṇa- in all of them makes these verbal stems thematic and puts all the roots of the list at the same level.

Similar remarks would hold true for the treatment of the former Sanskrit class 7 characterized by the varying nasal infix, -na- (strong grade) / -n- (weak grade), e.g. root chid- ‘to cut’, strong form chi-na-t-ti ‘he cuts’, weak form chi-n-d-anti ‘they cut’, in Aggavaṃsa's Pali grammar: the class is recognized (rudhādi)⁵⁷ but, since the defining variation does not exist in Pali where forms have been leveled, thus chindati ‘he cuts’, chindanti ‘they cut’, the contents are bound to change. Roots taken from the Sanskrit model are included with their new conjugation pattern (e.g., rudh ‘to obstruct’, yuj ‘to connect’, bhid ‘to split’ and a few others). But this new pattern, clearly identified through the presence of a consonant group including a nasal (-nda-, -nja-, -nca-) attracts a group of other verbs which would be classified in another category in Sanskrit lists: e.g. lip ‘to smear’, present limpati, sic ‘to sprinkle’, present sincati, etc. In Sanskrit these roots belong to class 6, unaccented thematic presents headed by tudati and including a number of roots strengthened by a penultimate nasal. In his own manner, Aggavaṃsa shows awareness of the objection which could be raised: why, then, are these verbs not included in the thematic class (Pa. bhuvādi) which also has verbs having a nasal? He embarks upon a lengthy discussion⁵⁸ arguing that one of the differences lies in the nominal forms: nouns derived from a verb of the thematic class keep their nasal (e.g. saṃkati ‘he suspects’, saṃkā ‘suspicion), whereas those derived from a verb of the class rundhādi do not retain the nasal (e.g. chindati ‘he cuts’, but cheda ‘cutting’). Other objections and answers advanced in this debate, interesting as they are, cannot be detailed here. The purpose is simply to show how the Pali grammarian negotiates between an extended model and the facts of the language he deals with.

Comparing the treatment of the verb ‘to conquer’ in grammars of Prakrit and Pali is interesting. Its inclusion in the Prakrit sūtras (Vararuci 8.56 and Hemacandra 4.241 quoted above) gives emphasis to the present form jiṇai, but Hemacandra's commentary also mentions jayai as optional. The latter comes from Skt. jayati, which is the common form of the present attested for this verb in the language since the Ṛgveda. It is likely that Pkt. jiṇai is a complete innovation, but it is also likely that the existence of the Skt. root jyā /jī ‘to overpower’, which has jināti for its present form may have encouraged such a form at some stage. The fact remains that a -na- present for this root seems to be typical of Middle Indian. Indeed, what we read in Hemacandra's Prakrit grammar in 12^th c. Western India for Prakrit, also applies to the two Pali grammars considered here, Kaccāyana and Aggavaṃsa, but only the latter enters into a critical discussion of the topic.⁵⁹ Aggavaṃsa first treats ji jaye ‘the root ji conveys the idea of victory’ as a member of class 1, the thematic present stems.⁶⁰ He therefore lists examples containing forms such as jayati, or its variant jeti. Nevertheless he ends by saying: ‘But this root has also forms such as jināti ‘conquers’ (present), jinitvā ‘having conquered’ (absolutive) when it goes to the group kiyādi’, i.e. the present class where the affix -nā- is inserted.⁶¹ So when he treats this class, he comes again to this root and unfolds a great variety of forms and derivatives based on jināti.⁶² Among them stands the substantive jina, a technical term designating the enlightened person possessed with omniscience in Buddhism,⁶³ and also in Jainism (derived from Jina precisely). In both traditions, this noun has been linked with the root ji-, but the end part -na can be explained only from a verb form containing a nasal. There are, then, two possibilities: jina either comes from jyā, jināti ‘to overpower’ (see above)⁶⁴ or has been encouraged by present forms jināti (late Sanskrit form instead of jayati) ‘to conquer’. Conversely, one feels that forms such as Pa. jināti (and Pkt. jiṇai) may have been favoured because they make the connection with a fundamental noun more obvious (than jayati). The illustrative sentence (identified as coming from the Vinaya) quoted in Kaccāyana remarkably shows the asserted etymological link between the substantive jina and the verb ‘to conquer’ used in various tenses:

pāpake akusale dhamme jināti, ajini, jinissatī ti jino ‘A Jina because he overcomes, has overcome and will overcome evil and bad things’. ⁶⁵

As we will just see, the importance attached to synchronical etymological links is not without consequence on grammar.

Keeping to the general Sanskrit frame for presenting and classifying verbal roots does not prevent Aggavaṃsa from introducing innovations. For instance, he has bhidi vidāraṇe ‘the root bhid- in the meaning to cut off’.⁶⁶ At first sight this is fully identical with what one reads in Sanskrit lists of roots: bhidira vidāraṇe.⁶⁷ But it is surprising to see that the phrases or quotations adduced as illustrations feature the word bhikkhu ‘monk’:

he is called bhikkhu because he cuts off the bad or wrong things,

a statement explained through a canonical quotation:

one is not a monk insofar as he seeks alms. The one who has expelled bad and good and lives in the world with discrimination, he is the one who is called bhikkhu.⁶⁸

According to linguistic etymology, the word bhikkhu is an agent-noun derived from the root bhikṣ- (Sanskrit form), bhikkh- (Pali equivalent) ‘to seek alms’ and refers to the monk as an individual who is dependent on the layfollowers for survival and support. In the Sanskrit lists of roots this root is mentioned as such: bhikṣa bhikṣāyām alābhe lābhe ca⁶⁹ ‘bhikṣ- in the meaning to seek alms whether they are obtained or not’. This root is also taught in our Pali grammar: bhikkha yācane.⁷⁰ ‘bhikṣ- in the meaning to ask for’, and the word bhikkhu is duly mentioned on this occasion but with no further analysis. Under these circumstances, the connection of Pali bhikkhu with the root bhid- and not with the root bhikṣ-/bhikkh- is clearly unexpected. But it is meaningful: in the non-Brahmanical traditions (Buddhism and Jainism), the so-called heterodoxies, we see that significant terms are reinterpreted on the ethical plane. Synonymic designations for ‘monk’ are among them. So we have so-called ‘popular’ or ‘folk’ etymologies that are not supported linguistically but convey fundamental values in given religious ideologies.⁷¹ What seems interesting is to see that they are given an additional legitimation through their inclusion in a grammatical work. This is an instance of how the traditional and overwhelming Sanskrit analytical frame is diverted and reappropriated in order to serve a different purpose. This is also in agreement with the initial purpose of the Pali grammar, which is to explain whatever was taught by the Buddha.

The limits of the Sanskrit model applied to the description of the Prakrit verbal lexemes, and the limits of the perception of Prakrits as being “derived” from Sanskrit emerge through the concept of ādeśa ‘substitute’ which finds a particular application in the field of verb roots or stems. It is used in case the grammarians feel that a given verb which is actually used in Prakrit cannot be traced back to Sanskrit through the application of the phonetic rules which edict regular phonetic changes between Sanskrit and Prakrit.⁷² Thus the concept of “substitute” in this particular understanding is a technique which helps including within a description inspired by the Sanskrit format of the Dhātupāṭha (root + meaning) the actual variety of Prakrit verbs since these lists include many that are not attested in Sanskrit sources but announce later linguistic evolutions.

In addition, it confines to semantics as the lists delineate the multiple shades of meaning gathering around a fundamental idea “to speak”, “to eat”, etc. which attract a rich verbal vocabulary. The most important source in this respect is the fourth section of Hemacandra's Prakrit grammar where a number of them are listed (4.1-259). ⁷³ These substitutes are optional or compulsory, depending on the indicatory letter found at the end of the root as we will see. The pattern is root (genitive case) followed by list of substitutes in the stem form collected together in the form of a nominative dvandva compound. The root is given in its Sanskrit form, with the final indicatory letter.

kather vajjara – pajjara – uppāla – pisuṇa – saṃgha – bolla – cava – jaṃpa – sīsa – sāhāḥ,

Of the root kath + i ‘to say’ (the 10 substitutes are): vajjara – pajjara – uppāla – pisuṇa – saṃgha – bolla – cava – jaṃpa – sīsa – sāha (4.2).

Since the root has the final i as indicator (katheḥ = genitive of kathi-), it means that beside all these forms, the form kahai, i.e. the Prakrit form obtained from Sanskrit kathayati through Prakrit phonetic rules prescribed earlier in the grammar is also possible. The grammarian is quite aware that the list of substitutes is somewhat heterogenous. Indeed, some of them can easily be traced back to Sanskrit, such as pisuṇai < Skt. piśunayati ‘to calumniate’ or jampai < Skt. jalpati which may have a slight derogatory meaning ‘to mutter’. But Hemacandra admits that some of them are regional words that are listed by other authors in collections of local words. Nevertheless, he adds, he considers them as root-substitutes so that they may receive the various affixes and produce thousand forms. The process of affixes introduced and removed, he concludes, applies to them as it does to Sanskrit roots.⁷⁴ If we analyse the substitutes, we see that the list results into a broad spectrum of shades of the basic idea “to say” and that some of the Prakrit verbs are effectively continued in modern Indo-Aryan languages.⁷⁵

4 Concluding remarks

The perception of Pali and Prakrit in the respective grammars of these languages may explain why the language used for description is Pali in one case and Sanskrit in the other. The Pali and Prakrit grammars show how technical categories and terms may have been transferred from Sanskrit models. This is no hindrance to the coinage of new terms, albeit marginally (Pali puthuvacana ‘plural’ in the Saddanīti). The metalanguage used seems to be of limited impact on the descriptions of the verb as evidenced by the Prakrit grammarians. The fact that they use Sanskrit and consider Prakrit as ‘derived’ from Sanskrit does not prevent them from describing features that are synchronically specific to Prakrit and distinct from Sanskrit. The situation shown by the Pali grammarians is slightly more complex in this respect: although they use Pali, and not Sanskrit, they acknowledge the existence of fundamental classifications used in the Sanskrit grammatical tradition. But their foremost purpose being the elucidation of the Buddha's word, and their explicit source being the scriptures reflecting this teaching, which are profusely quoted, their description of the verb adjusts to the new linguistic reality that Pali represents. Thus, the grammars of the Middle Indian languages, whether they use Pali or Sanskrit as their descriptive languages, are far from being limited to rules of mechanical substitution although they seem to be prevalent at first sight.

Abbreviations

Kacc: see below: Kaccāyana

MIA =: Middle Indo-Aryan

Pa. =: Pali

Pkt. =: Prakrit

Sadd: see below: Saddanīti

Skt. =: Sanskrit

Primary sources

Secondary sources