Pāli and the languages of early Buddhism

This article goes back to the beginnings of Buddhism, in fact to that ancient problem: What language or languages did the Buddha speak? It discusses Old Indo-Aryan and the origin of the Prakrit dialects, the language(s) of the Buddha and the Jina, Old Māgadhī, the migration of the Buddhist teachings, the writing down of the canon in Ceylon, anomalous forms in the Theravādin canon, anomalous forms and the Aśokan inscriptions, and why anomalies remain in the Pāli canon.
 I want, then, to go back to the beginnings of Buddhism—in fact to that ancient problem: What language or languages did the Buddha speak? I might as well start by admitting that we do not know, but we can make some intelligent guesses. Having made those intelligent guesses, I should then like to say something about the relationship between that language or those languages and the language of the Theravādin canon which we call, traditionally if incorrectly, Pāli.

Of the various dates which have been put forward, I start from the premise that Indo-Aryan speakers started to come into India from the West somewhere around the beginning of the second millennium B.C.E. It is probable that they moved into India in waves, and since we can assume that Indo-Aryan was developing all the time, it is likely that, if there was a time span of several generations between incursions,each wave spoke a slightly different dialect.

As they moved into India, they came into contact with other peoples who were already settled there. Some of these, who had also come into India from the West, perhaps a thousand years earlier, spoke Dravidian languages and others, further East, spoke Munda languages. They also met speakers of another language, of which we know nothing except that, as might be expected, its vocabulary included terms for agricultural features, trees, fruits, animals, etc. for which the IndoAryan language had no equivalents. From all of these sources Indo-Aryan borrowed vocabulary, and possibly phonological features too. It seems very possible that there would have been conflict of some sort between the incomers and the indigenous inhabitants, but it is also likely that, after that initial conflict, the races would have begun to assimilate, and there would have been an interaction between their various languages.

We know from the differences between the R̥gvedic language and later Sanskrit that there were phonological and morphological variations in the Old Indo-Aryan dialects. We also find in the R̥gveda a number of forms which our knowledge of Indo-European philology tells us we should not expect in early Sanskrit. We can regard them as ‘Prakritisms’, i.e. early indications of the changes which distinguish Middle Indo-Aryan from Old Indo-Aryan.

Even though the recension of the R̥gveda which has been handed down to us is perhaps no earlier than the seventh century B.C.E., nevertheless we can see that such proto-Middle Indo-Aryan features were in evidence well before the time of the Buddha. If there were such features in the literary language of the R̥gveda, the assumption is that the spoken languages of the ordinary people had many more of them. We can, in fact, deduce that their languages contained not only morphological and phonological differences from Sanskrit, but also differences of vocabulary, and variations in the way in which words were formed. Nevertheless, despite these indications of unexpected features occurring at that time, we would date the full development of the Middle Indo-Aryan dialects some centuries later.

The language(s) of the Buddha and the Jina

We can be certain that those dialects had developed by about 400 B.C.E., because by that date they were sufficiently wide-spread to present a threat to the predominance of Sanskrit. It seems clear that Pāṇini and the other Sanskrit grammarians wrote their grammars in an attempt to define and fix the nature of Sanskrit, so that the inroads of Middle Indo-Aryan into the sacred language could be identified and checked.

We read of statements by the Buddha about regional differences of vocabulary. We can therefore conclude that there were regional dialects in use in different parts of India at his time. I assume that the Buddha’s travels were sufficiently widespread to have taken him from one dialect area to another.

But can we be certain that the Buddha did not preach in Sanskrit, and that translations of his teachings from Sanskrit into Middle Indo-Aryan were not made later?

Any movement, whether religious or social, which opposed the power and status of the brāhmaṇa caste might be expected to make use of a non-brahmanical language. Not only would this be an anti-brahmanical gesture, but it would also aid the non-Sanskrit-speaking element of the population, i.e. those who spoke a vernacular language. It therefore seems very likely that the Buddha’s sermons were preached in a non-Sanskritic language.

There are parallel statements in Jain and Buddhist texts about the nature of the languages which the founders of Jainism and Buddhism employed. Both religions stated that those languages could be understood by every living creature, humans and animals. I noted that the similarity between the Jain and Buddhist statements is too great for them to be unconnected.

It is also clear that the Jain and Buddhist texts go back to a common idea about the development of language, for they explain that the languages of the Jina and the Buddha respectively were intelligible to all beings because from them developed the other languages, i.e. they were the root languages. It is not difficult to see that this theory was directly opposed to brahmanical ideas about Sanskrit, the exclusive property of the brāhmaṇa caste, and by the 5th century B.C.E. all but unintelligible to the lower castes, let alone animals.

In short, we can see that Gotama the Buddha and Mahāvīra the Jina, living as they did in closely related areas in eastern India, shared a common Indian cultural context, with a common theory of language. The similarities in religious terminology in Buddhism and Jainism show very clearly that the two religions arose from a common religious milieu, united in their opposition to brahmanical ideas. It seems, then, more than likely that in their rejection of the religion of the priestly brāhmaṇa caste they also reacted against the brahmanical language and made use of the vernacular dialects spoken by the non-brahmanical castes, which we may assume were closely related dialects of an eastern Middle Indo-Aryan language.

The texts tell us that the Buddha preached in various places, and we can assume that, as he moved about, he varied his language to suit his audiences. Much of his teaching life was spent in Magadha. We can therefore deduce that on some occasions, at least, he used the Magadhan dialect of the time, which we may call Old Māgadhī.

We have little knowledge about the dialect geography of the fifth century B.C.E. To help us in our attempt to identify Old Māgadhī we have two main aids: the inscriptions found in India, from the time of Aśoka onwards, and the statements of the Indian grammarians. Both sources refer to a time later than the Buddha, and are to some extent unreliable.

Many scholars now believe that the Buddha’s death is to be dated c. 400 B. C.E. This means that the Aśokan inscriptions are about 150 years later, and so we must extrapolate backwards to find the situation at the time of the Buddha. It is clear from the Aśokan inscriptions that different areas had different dialects, and the divergence is such that we can suppose that the differences had existed for a considerable time. Although we can surmise that Aśoka intended his inscriptions to be carved in the dialects appropriate to each site, there is clear evidence that this was not always done. We have, then, to accept that the Aśokan inscriptions cannot be entirely satisfactory as a guide to earlier dialects and dialect forms.

The grammarians were writing many centuries later, from the fifth century C. E. onwards, and were in any case describing the literary forms of the dialects. Nevertheless, a literary dialect or language very often includes fossilised features which were used at a much earlier date.

As a third aid, there is the information which can be gained from Jain texts, which underwent the same type of language changes as the Buddhist texts, as Jainism spread. Attempts have been made to identify Old Ardha-Māgadhī, the earliest form of the language of the Jain canon, and we can, to some extent, use those studies to help with the identification of the languages of early Buddhism.

Most useful of all is the inscriptional evidence that by historical times a dialect was being used in Magadha which turned all -r- and -l- sounds into -l-, turned all three sibilants into the palatal ś, and had the nominative singular of short a-stem nouns in -e. Its speakers seemed to have remained close together, for the epigraphical evidence suggests that its use was restricted to a very small area, and this set of features is not found together in any single Aśokan site.

Since the characteristics of this inscriptional dialect agree closely with the features of Māgadhī, as described by the grammarians many centuries later, it would seem that Māgadhī was a very conservative dialect, changing little over a long period of time. Extrapolating on this basis, we can perhaps be fairly confident that the main features of Old Māgadhī at the time of the Buddha were those of the inscriptions I have just mentioned.

As Buddhism moved from the land of its origin into areas where different dialects or languages were spoken, and as those dialects or languages developed and changed over the course of time, it would seem to be inevitable that some sort of translation process was needed if the Buddha’s teachings were not to become unintelligible. We must remember that at this time those teachings were still in an oral form, which meant that a preacher was not bound by the wording of a rigid written canon, but could translate (or transpose, if translate is thought to be an inappropriate term) freely as the need arose.

We know nothing about the translation techniques which were employed by those who took the Buddha’s teachings to an area where a different dialect was used, and who were therefore faced with the task of translating into that dialect. If the dialect from which the translation was being made and the dialect into which the translation was being made had differences of phonology and morphology, then much of the translation process could have been done almost mechanically. For example, if the translation was being made from a dialect which did not voice intervocalic consonants into one which did voice them, than the translator simply had to voice all unvoiced intervocalic consonants. If the translation was being made from a dialect where the nominative singular of short a-stem nouns was in -e into a dialect where it was in -o, then the translator had simply to change all nominative singular -e endings into -o.

The accuracy of such a translation process depended on the knowledge and ability of the translator. Clearly, he had to know something about the characteristics of the donor and receiving dialects. If, by any chance, the donor dialect was one which voiced intervocalic consonants, and the receiving dialect did not voice them, then a mechanical translation technique was likely to produce errors, because it would lead to a situation where consonants which should be voiced in the receiving dialect might become unvoiced. It was all very well changing all past participles in -ida in the donor dialect into -ita in the receiving dialect, but what about a form like uppāda? Since there is a root pad- as well as a root pat-, should this be translated into uppāda, or left asuppāda? Only the sense of the passage could determine this, and if the passage was ambiguous, or for some other reason was difficult to understand, then a translator was left to guess, and his guesses might not always be correct. If the receiving dialect had the nominative singular of short a-stem nouns in -o, then to turn all -eendings into -o may well have meant that forms which should have had the -e ending in the receiving dialect, e.g. locative forms, were also changed, incorrectly, into -o.

According to the tradition of the Buddhist chronicles, Theravādin Buddhism was introduced into Ceylon (Sri Lanka) at the time of Aśoka around the middle of the third century B.C.E. We know nothing of the form of the language which was used by the Buddhists at that time. We know only that, according to the same tradition, the canon of their texts was written down in the first century B.C.E., i.e. about 200 years later.

Little further development in the language of the canon seems to have taken place after that date. Very few changes can be pointed out which are to be ascribed to the influence of the Sinhalese Prakrit, and even those few are debatable. There are some Sanskritisms, and some influence from the twelfth century grammarians has been detected,but we can now, within limits, define the influence which the later commentators and grammarians had upon the canonical language, and identify many of the features which their rules imposed upon it.Having done that, we can say with certainty that the basic language of that canon is not Māgadhī. If we are correct in believing that the Buddha spoke Old Māgadhī on occasions, then we can see that the texts have been translated at least once before they were written down.

Is there any evidence for this? Yes, there is. If we analyse the language of the Theravādin canon, we can see that for the most part it has features which we would class as western, following the standard analysis of the languages of the Aśokan inscriptional sites. Nevertheless, we can detect anomalous forms in it, i.e. forms which do not seem to follow western patterns of phonology and morphology.

In what I say in this paper I am, to some extent, a prisoner of my own nomenclature. I use ‘Pāli’ as the name of a language, and it has become conventional to do so. But Pāli is an abbreviation of pāli-bhāsā, which means the language of the pāli, i.e. the texts, or the canon. It follows, then, that every feature of the language of those texts, however strange, and however inconsistent with the rest of the language of the texts, is Pāli. We can define it, if we wish, as a western dialect with some eastern features, but it would, technically, be incorrect to talk about these possible eastern features as anomalous forms. It is even more incorrect to call them non-Pāli forms. Nevertheless, I must call them something in the course of my discussion, so I shall call them ‘anomalous’.

The presence of these anomalous forms in the Theravādin canon was, of course, noticed a long time ago, and their value as indicators of an earlier form of the Buddha’s teachings has been much studied and debated. Lévi saw them as précanonique (‘pre-canonical’), while Lüders regarded them as evidence of an Urkanon (‘an original canon’).

If these anomalous forms are remnants of dialects through which the Buddha’s teachings, or portions of them, were transmitted, then our task is to see whether they exhibit characteristic features of any of the dialects about which we have knowledge, in the hope that this will tell us about the regions and languages through which the texts, or parts of them, were transmitted.

Within the scope of this paper I can consider only a small selection from the evidence which has been collected:

⁰Anomalous phonological features

1 The voicing of intervocalic consonants, e.g. -t- 〉 -d- niyyādeti (〈 niryātayati), saṅghādisesa (〈saṅghātiśeṣa), [opapātika (〈 *aupapādika)].

2 The voicing of doubled intervocalic retroflex consonants, e.g. -ṭṭh- 〉 -ḍḍh-: kaḍḍhati (〈 kaṭṭha 〈 kr̥ṣṭa), aḍḍhuḍḍha (〈 ardhacaturtha).

3 The development of intervocalic consonants, e.g. -k-, -j-, -t-, 〉 -y-: Sabhiya (〈 Sabhika), niya (〈 nija), Māyā (〈 Mātā).

4 The development of intervocalic -y- to -j-: gavaja (〈 gavaya).

5 The development of initial y- 〉 j-: jantāghara (〈 yantra-gr̥ha), je (〈 ye), [Yamataggi (〈 Jamadagni)].

6 The change of retroflex -ṅ- 〉 dental -n-: nibbāna (〈 nirvāṇa), bhikkhunī (〈 bhikṣuṇī), bhūnahū (〈 bhrūṇahan).

7 An alternation between -l- and-r-: pali- (〈 pari-).

8 An alternation between intervocalic -y- and -v-: āvuso (〈 āyuṣ-), āvudha (〈 ādyudha).

9 The development of -kṣ- 〉 -kkh- instead of the expected -cch-: bhikkhu.

10 The development of a nasal + an unvoiced consonant to a nasal + a voiced consonant: handa (〈 hanta), pārājika (〈 *pārañjika 〈 pārañcika).

11 The development of a nasal + a voiced consonant to a doubled nasal consonant: Channa (〈 Chanda), ārammana (〈 ālambana).

12 The contraction of -ā̆ya- 〉 -ā- instead of -e-: sotthāna (〈 svastyayana), pācittiya (〈 prāyaścitta).

13 The resolution of consonant groups instead of assimilation: magga-jina (〈 mārga-jña).

Anomalous morphological features

14 -e as the nominative singular of short a-stem nouns.

15 -āse (〈 Vedic -āsas) as the nominative plural of short a-stem nouns.

8. Anomalous forms and the Aśokan inscriptions

I should like now to examine some of these anomalous forms, to see whether it is possible to put them into context with datable language features more accurately (p.142) than was possible twenty years ago. This will help us to see how much earlier they may be than the first century B.C.E., and to deduce in what part of India they originated.

Phonology

1 We find voicing of -t- 〉 -d-, e.g. in niyyādeti (〈 niryātayati ‘he assigns’), and sa ghddisesa (〈 saṅghdtiseṣa ‘a class of offence’), and such hyper-forms as opapātika (〈 Sanskrit *aupapādika ‘self-produced’) show that the translator had knowledge of a dialect where voicing occurred. Voicing is not a consistent characteristic of any of the Aśokan dialects, but there are some anomalous voiced forms in the Aśokan inscriptions, e.g. loga 〈 loka, ájala 〈 acala, Ubi 〈 Upi, thuba 〈 thūpa, and the hyper-forms which occur, e.g. the root pat- written for pad-, show that the Aśokan scribes had knowledge of a dialect where voicing occurred.

2 The voicing of -ṭṭh- 〉 -ḍḍh- occurs in kaḍḍhati (the denominative verb from kaṭṭha 〈 kr̥ṣṭa ‘dragged’) andaḍḍhuḍḍha (〈 ardha-caturtha ‘three and a half). There is an anomalous example in Aśokan a(ḍ)ḍha- (〈 aṣṭau ‘eight’).

3a There is the change of intervocalic -k 〉 -y- in Sabhiya 〈 Sabhika, in the name of the Sabhiya-sutta of the Pāli Sutta-nipāta. The change of intervocalic consonants to -y- is a characteristic of Māhārāṣṭrī Prakrit, but at a later date than the Aśokan inscriptions, where there are only one or two examples, some of which can probably be explained otherwise.The -ikya forms found at the Kālsī site, where the other sites have -ika, may, however, reflect some such change. It is possible that the scribe was writing the -ky- ligature because he received -k- in his exemplar, but wanted to show that the local pronunciation was nearer -y-. Lüders has given some examples of the -ika/-iya alternation, and the fact that the change of -k- 〉 -y- occurs after the vowel -/- seems to support the view that -ikya at Kālsī does indicate that the change 〉 -iya had taken place or was beginning to take place there.

3b It is probable that niya 〈 Sanskrit nija ‘own, belonging to oneself’, is an example of the change of intervocalic -j-〉 -y-. It has been shown that some of the etymologies in the Sabhiya-sutta depend upon their being first pronounced, i.e. composed in, not just transmitted through, a dialect where some of the intervocalic consonants, including -j- had developed into -y-. The fact that the sutta is an early composition is shown by the fact that there is a Buddhist Hybrid (p.143) Sanskrit version of it in the Mahāvastu, and if the age of the text supports the view that the version underlying both the Pāli and the Buddhist Hybrid Sanskrit versions was composed in Magadha, then this is additional evidence for the existence of this sound change in that area at an early date.

3c The Sabhiya-sutta also has etymologies which depend on their being pronounced in a dialect where intervocalic -t- had developed into -y-. This would support the suggestion that the Buddha’s mother Māyā was probably not called ‘Delusion’ but ‘Mother’ (Sanskrit Mātā). There is codasa (〈 *ca(y)uddasa 〈 caturdaśa ‘14’) in an eastern Aśokan inscription.

4 We find intervocalic -y- 〉 -j- in gavaja, which occurs instead of the expected gavaya (‘a type of ox’). This may be a hyper-form, but it may be a genuine remnant. This change is not typical of the Aśokan inscriptions, but it has been noted that there is one occurrence of an intervocalic -y- becoming -j- there, in the word for ‘peacock’ (Sanskrit mayüra). It occurs as majūla at the Kālsī and Jaugada sites, with the eastern l replacing the r, and as majura in Shāhbāzgaṛhī and Mānsehrā in the North-west, with the expected western r. The appearance of -j- at four sites suggests very strongly that -j- was taken over without correction by the scribes at all those sites from the exemplars which they received. The dialect at Shāhbāzgarhī has a tendency to turn intervocalic -j- into -y-, and would therefore be unlikely to do the opposite, unless there was a good reason for doing so.

It is to be noted that this deduction, made on the basis of one single word, implies that the dialect of the exemplar (which was perhaps Aśoka’s own dialect) differed from the eastern dialect attested in his inscriptions. This implication need present no difficulties, because there is also evidence that Aśoka’s own dialect had palatal š for the sibilant,whereas no version of the Aśokan inscriptions, other than those from the North-west, has this sound except by scribal idiosyncrasy.

5 There is the change of initial y- 〉 j- in jantāghara (〈 yantra-gr̥ha ‘bathhouse’) and the particle je used when addressing women of a lower class.This particle seems to be related to the emphatic particle ye which appears in Pāli and the Aśokan inscriptions. We may compare haṃje, which is used in Jain Prakrit in a very similar sense to je. From its wide-spread appearance in the Ardha-Māgadhī of the Jain canonical texts, we may assume that the change of initial y〉 j- was a feature of Old Ardha-Māgadhī.

There is also the hyper-form Yamataggi for the Sanskrit name Jamadagni. We can deduce that the translator was aware of the fact that the donor dialect sometimes showed initial j- where the receiving dialect showed y-, and intervocalic -d-where the receiving dialect showed -t-. When he came across the form Jamadagni (or more likely Jamadaggi), in the absence of any knowledge of the correct form of the name in the receiving dialect, he was obliged to back-form by rule, which led to the hyper-form. This confirms that there was a pre-Pāli dialect where the changes of initial y- 〉 j- and intervocalic -t- 〉 -d- occurred.

6 There is the change of retroflex -ṇ- 〉 dental -n- in nibbāna, bhikkhunī, bhūnahū (〈 bhrūṇahan ‘embryo killer’). This is a standard feature in the eastern versions of the Aśokan inscriptions.

7 We find -l- instead of -r- in, e.g., verbs with the prefix pali- instead of pari-. This is a standard feature in the eastern versions of the Aśokan inscriptions.

8 Intervocalic -v- occurs instead of -y- in āvuso (〈 āyuṣ- ‘sir’) and āvudha (〈 āyudha ‘weapon’). This alternation (particularly in the usage of -v- and -y- as glide-consonants) is common in Aśoka, e.g. the optative endings -evu in the eastern versions and -eyu in the western versions.

9 We find -kṣ- developing 〉 -kkh- instead of -cch- in bhikkhu. This is a standard feature of the eastern versions of the Aśokan inscriptions.

10 There is a very small number of forms which show an anomalous development of a consonant group containing a nasal + an unvoiced consonant: -nt- 〉 -nd-: the vocative particle handa 〈 hanta. I accept the suggestion that pārājika (‘a type of offence’) is 〈 *pārañjika 〈 pārañcika.

11 There is also a very small number of forms which show an anomalous development of a consonant group containing a nasal + a voiced consonant: -nd- 〉 -nn-: the Buddha’s charioteer name Chanda changing 〉Channa; -mb- 〉 -mm-: ālambana ‘support’ changing 〉 ārammana.

There is an example of the second change in Aśokan Luṃmini 〈 Lummbini. Both changes are typical of the Gāndhārī dialect as seen in the Gāndhārī Dharmapada, but they are not found in the version of the Gāndhārī dialect which occurs in the Aśokan inscriptions in the North-west. They therefore seem to be a later development in that dialect, and it is most unlikely that these forms could be borrowings into Pāli from the Gāndhārī dialect.

12 Instead of the usual contraction of -aya- 〉 -e-, we find -ā̆ya- contracting 〉 -ā- in, e.g. sotthāna (〈 svastyayana ‘welfare’), pācittiya (〈prāyaścitta ‘expiation’). This change is found in Jain texts.

13 There are resolved consonant groups, e.g. -jñ- resolved 〉 -jin- as well as assimilated 〉 -ññ-, in magga-jina (〈 mārga-jña ‘knowing the road’). The resolution of consonant groups is an eastern feature in Aśoka, with assimilation in the West.

Morphology

14 The masculine nominative singular ending -e in place of -o is perhaps the most widely attested of the anomalous forms. It is a standard feature of the eastern versions of the Aśokan inscriptions.

15 The nominative plural ending -āse occurs spasmodically in Pāli. It is found in the eastern dialect of Aśoka.

 Dialect variation

We may, then, deduce that at the time of Aśoka, and even earlier in the case of some Pāli texts which we can, with great probability, date to a pre-Aśokan time, many of the features which we have marked out as anomalous in Pāli can be shown to occur in eastern versions of Aśoka’s inscriptions, although sometimes only as a single example, i.e. they are anomalies in Aśoka also.

We must therefore conclude that the information we have hitherto had about dialects in Aśoka’s time, let alone at an earlier time, is deficient, since it by no means tells us about all the dialects which were in use. The fact that we can, if we look carefully, augment this information by detecting hints of other dialects is, in itself, of great importance, because it suggests that the anomalous forms in the eastern versions of the Aśokan inscriptions, e.g. -y- 〉 -j-, may be traces of the dialect which Aśoka himself used, rather than the dialect of Aśoka’s secretariat at Pātaliputra. Aśoka probably spoke a sub-dialect of Māgadhī, i.e. a dialect spoken in just the area where we would expect the early language of Buddhism to be in use. That dialect is not represented, in its entirety, in any extant version of his inscriptions.

If our conclusion is correct, then, we can assume that the dialects in use, even in a limited area such as Magadha, showed variations. This should not surprise us, because we should expect neighbouring areas, even neighbouring villages, to have very slightly divergent dialects. We can also assume that traces of this linguistic diversity were retained when the Buddha’s sermons, which had been preached in different areas, were first collected together and their language was homogenised. We can guess—and it is nothing more than a guess—that when the first collection of the Buddha’s teachings was made and homogenisation began to take place, examples of divergence were even more numerous, since, as sub-dialects of the Magadha area, these variations were probably not sufficiently great to cause difficulties for speakers of other sub-dialects, and there was therefore no need to remove them.

Sometimes these anomalous forms that I have been discussing are called ‘archaisms’ in as much as they are examples of earlier features which are no longer standard in Pāli. Some call these forms ‘Māgadhisms’,because they conform in part to the pattern of Māgadhī, as described by the grammarians. Others, however, say that the term Māgadhism is misleading, because it takes for granted that these forms are taken over from Māgadhī.It is true that it is not entirely accurate to regard all such anomalous forms as Māgadhisms, since some of these features are found in other dialects too—the -e nominative singular ending, for example, is found in the post-Aśokan Kharosthī inscriptions from north-western India.

It has been pointed out that other explanations can be given for some of these forms, but there is unlikely to be agreement about such views, since there is no way of proving or disproving any hypothesis. What is, however, clear is that every Māgadhism does not prove that the passage in which the Māgadhī form occurs is old and dates from the Māgadhī period of Buddhism, and certainly those who say that we can rely too much on such details are correct. Once a standard procedure had been adopted, e.g. of writing the ‘Māgadhī’ vocative plural form bhikkhave in a specific context, it was also applied to newly created texts, so that the occurrence of such Māgadhisms tells us nothing about the original language of the text in question. In short, the occurrence of a Māgadhism in a text does not prove that the text was originally in the Māgadhī dialect.

 Reasons for retention

Irrespective of the name given to them, the question is, why did these anomalies remain in the Pāli canon? Why were these features overlooked when the general task of translating from one dialect to another was being carried out, as Buddhism moved away from its eastern home, and—most importantly—at the time when the language of the canon was codified into the language which we call Pāli. The problem is not so much ‘where did they come from?’ as ‘why did they remain?’

An easy answer is ‘scribal error’, and we can point to the Aśokan inscriptions, where some very obvious errors of ‘translation’ can be seen, resulting from the need to translate quickly before they were inscribed or even while they were being inscribed. It is hard to imagine that this could apply to texts which were recited repeatedly over a period of some hundreds of years and were, we may surmise, subjected to a great deal of editorial ‘correction’ or ‘normalisation’ at the time the canon was written down.

There was, I think, no one reason for the failure to translate, but it is possible to suggest why some of the necessary changes were not made. If we examine the examples I have given we can surmise that saṅhādisesa, pārājika and pācittiya were retained because they are technical terms, being the names of particular types of offence; nibbāna, bhikkhu and bhikkhunī are also technical terms; bhūnahū is a semi-technical term andāvuso has a semi-technical sense, in as much as the Buddha laid down the specific circumstances in which it was to be used; jantāghara too has a semi-technical meaning as part of a vikāra.

The derivation of some words was probably not recognised, e.g. kaḍḍhati and āvudha, and translators were therefore unable to identify the correct form in their own dialects. The particle je was doubtless retained in its eastern form because the genuine western form ye was not used as a vocative particle in the dialect into which the sermons were translated, although ye is used as an emphatic particle after infinitives in Pāli in exactly the same way that je is used in Jain texts.

There were genuine forms with -y-, e.g. the suffix -ī̆ya, and in resolved consonant groups containing -y-, so it is possible that forms with intervocalic -y-, such as niya, seemed not to require translation.

Other forms were retained because their eastern form could be confused with another western form, e.g. -jina 〈 -jña was retained because jina ‘conqueror’ made sense in the context.

The same is true of the ending -e, which was sometimes kept because it seemed to make sense as a locative, e.g.ante-vāsin ‘one living at the anta’, rather than ‘one living inside (antar)’; pure, as a parallel to pubbe.

If a translator had to deal with a passage where a form was ambiguous, he might be unable to determine the correct form. For example, in one verse in the Theragāthā there are three words ending in -e:sacce, atthe and dhamme. These can be eastern nominative or western locative case forms. The Pāli commentary took all three as locatives, while the Buddhist Hybrid Sanskrit translator took one of them,sacce, as a nominative, and wrote satyam.

Some -e forms seem to be quotations. The Dīgha-nikāya describes the teachings of the six teachers who were contemporary with the Buddha, and some of these descriptions include nominative singular forms in -e. It seems likely that the words attributed to the six teachers reflect the actual dialects they spoke. Although not everyone would agree with this suggestion, it is supported by the fact that comparable views ascribed to heretical teachers in Jain texts show close verbal similarities.

We can surmise that the ending -āse was probably retained for metrical reasons. In verse passages replacement would present difficulties, since the standard nominative plural ending -ā would leave the pāda a syllable short.

Conclusions

We can conclude, then, that Lévi was to some extent correct. The technical and semi-technical terms are pre-canonical in the sense that they are earlier than the Theravādin canon, and in fact probably predate any form of Buddhist canon in that they go back to the very beginning of Buddhism. Since some of them have parallels in Jain texts we can surmise that they were part of the pre-Buddhist and pre-Jain šramaṇical terminology which was taken over by both Buddhists and Jains from the eastern dialect spoken in the area where both religions had their origin.

Lüders ‘suggestion, however, is less acceptable. Almost all the features of the other anomalous forms can be shown, or can be surmised, to have been current in the East at the time of Aśoka, and probably earlier. If we could be bold enough to say ‘150 years earlier’, then we could go further and say ‘employed in the East at the time of the Buddha’. We could be even bolder and say, ‘perhaps by the Buddha himself’, i.e. these features may have formed part of some of the very earliest texts. They may indeed go back to the original language or languages of Buddhism, but whether those texts were already grouped together into a systematised collection which could be described as an Urkanon is much more debatable. In fact, if the term Urkanon implies a single collection, in a unified language, approximating to the structure and contents of later canons, and accepted by all Buddhists before the different schools began to separate, then it would seem highly doubtful that such a thing could ever have existed.