On the Pronunciation of Jña

One of the pleasures of learning Sanskrit for an English speaker is that it is written almost exactly as it is pronounced. I say almost, because there are one or two irregularities. The most striking irregularity is probably the consonant cluster in jña. This comes out variously as gya, gnya, dya, dnya, and nya and, at face value, there is no rime or reason for this. However, it turns out to be a regional feature influenced by how languages deal with Sanskrit loan words.

In a Buddhist context the word jñā is fairly common as it occurs in technical terms like vijñāna, saṃjñā, and prajñā as well as more general terms like jñāna and jñātavyam. Prajñāpāramitā texts use words deriving from the root √jñā very often. The meaning of jñā is relatively straight forward; it means: "knowledge" (noun) and "to know" (verb). But when one interacts with Sanskritists, one realises that jña is pronounced in a number of more or less counter-intuitive ways, none of which reflect the written word. 

In this essay I will try to explain where these differences come from and try to identify how jña should be pronounced. I will argue that it should, like the rest of the language, be pronounced as written. My evidence will come from two main sources the reconstructed historical phonology of Sanskrit and the historical writing of Sanskrit.

We know that Sanskrit is part of the Indo-European family, an idea first proposed by William Jones in his famous address to the Royal Asiatic Society in 1786:

"The Sanskrit language whatever be its antiquity, is of a wonderful structure; more perfect than the Greek, more copious than the Latin, and more exquisitely refined than either, yet bearing to both of them a stronger affinity, both in the roots of verbs and in the forms of grammar, than could possibly have been produced by accident; so strong, indeed, that no philosopher could examine them all three, without believing them to have sprung from some common source, which, perhaps, no longer exists."

Note that the relationship is primarily based on similarities of grammar (the conjugations of the verbs and declensions of nouns), but one may also see it in cognate words, such as English know, Greek γνῶσις (gnōsis), Avestan: zan- (v. to know), and so on. Comparative linguistics allows us to compare existing languages and reconstruct the mother tongue of both. In the case of Sanskrit we can trace it back to a common root with Iranian languages, called Proto-Indo-Iranian (PII) , and then one step further to the common root Indo-Iranian and European languages called Proto-Indo-European (PIE). The reconstructed phonology has a logic to it and is internally consistent but no recordings or documents of any kind exist. All that exists are daughter languages.

Phonology

Articulation

We think we have a fairly accurate idea of the phonology of attested ancient languages in the region of Iran and North Western India: Vedic, Old Persian, and Avestan. And we can match this to information about articulation - the physical motions and points of contact of the tongue, mouth, and throat that produce vocal sounds. We can also determine qualities such as whether the motion stopped the air flow completely (if briefly) or only partially. A consonantal stop or affricate requires the tongue to completely block the air flow momentarily, i.e. Sanskrit k g ṭ ḍ t d. A fricative does not completely block the air, i.e. Sanskrit j c v ś ṣ s h and creates a turbulent air flow that changes the timbre of the sound. There is also an approximant which constricts the airflow but not enough to create turbulence.

In addition, a vocal sound may be accompanied by vibrations of the vocal cords or not, which we call voiced (Skt: g j ḍ d b) and unvoiced (Skt: k c ṭ t p). Or the sound may resonate mainly in the nasal cavity which we call nasal (Skt: ṅ ñ ṇ n m).

The /j/ and /ñ/ sounds of Sanskrit are made with the blade of the tongue pressed against the hard palate and are thus called palatal (Skt: c ch j jh ñ). Further forward is the alveolar region which is where the palate merges into the gums. English speakers pronounce /t/ and /d/ by contact of the tip of the tongue on the alveolar. Sanskrit distinguishes a true dental, i.e. tongue touches the teeth (t th d dh n) and a retroflex, i.e. tongue curls back so the tip touches the edge of the hard palate (Skt: ṭ ṭh ḍ ḍha ṇ). Indians tend to hear our English alveolar consonants as retroflex. So "doctor" becomes डक्टर् ḍakṭar. When the back of the tongue contacts the soft palate we call these sound velar region (Skt: k kh g gh ṅ). 

Historical Phonology

The reconstructed articulation of PIE is largely settled, though there are some controversial aspects to it. Fortunately, we don't need to consider the controversial parts. In order to explain the palatal consonants in Vedic, Old Persian, and Avestan languages we need to postulate three palatals in PIE that in fact were pronounced somewhere between the palatal and velar: (1) an unvoiced palatovelar affricate, /k̂/ (2) a voiced palatovelar affricate, /ĝ/ and (3) an aspirated voiced palatovelar affricate /ĝʰ/ (Burrow 1973). This notation is indicative of articulation rather than sound. As I understand it we cannot give precise pronunciations of these using the International Phonetic alphabet (I've found no sources that use IPA to notate PIE phonology).

To get to Sanskrit these sounds underwent two parallel processes of "palatalisation". In most cases these consonants drifted towards the front of the mouth slightly and became fricative. In the first palatalisation (using the notation from Burrow 1973) PIE: k̂ ĝ ĝʰ first became Proto-Indo-Iranian: ć ȷ́ ȷ́ʰ and then ś ź źʰ. At this point the PII language family started to break up.

In the Avestan form of Old Iranian these sounds became s z z. In the Persian form of Old Iranian ś and źʰ became θ and z, while from the earlier form ȷ́ we get (directly) d. Similarly in Vedic ś and źʰ become ś and h, while from ȷ́ we get j. We also see j deriving another way, which is evident in some sandhi differences, but exploring this would take me too far from my aim to explore jñā. The j in jñā derived this way.

Sanskrit jñā is cognate with Avestan zan- and Old Persian dān-. Note the insertion of a vowel between the initial consonant and the nasal. This is something we also see in practice in Sanskrit. The verbal form of the root jñā is jānāti, though we also see jñāta, jñātum, jñātva, and jñāyate.

We can now start to put some more precise values on these sounds using the International Phonetic Alphabet (IPA is conventionally put in square brackets). The "palatals" in Sanskrit are c ch j jh ñ, but in IPA notation are actually alveolo-palatal and in IPA are written: [tɕ] [tɕʰ] [dʑ] [dʑʱ] [ɲ]

If you want to know how these sound, cut and paste the symbol into a search on YouTube. Each one is demonstrated in a short video

Here we are focused on the voiced palatals.

[ʒ] voiced post-alveolar fricative
[dʒ] voiced post-alveolar affricate
[dʑ] voiced alveolo-palatal affricate
[dʑʱ] breathy voiced alveolo-palatal affricate
[j]   voiced palatal approximant
[ɟ]   voiced palatal affricate
[ɲ]  voiced palatal nasal
[nʲ] or [ɲ̟] voiced palatised nasal

We would expect jña to be pronounced [dʑɲɐ] or [ɟɲɐ] (most sources seem to prefer the latter). As noted above, jña is almost never pronounced this way. Ashok Aklujkar points out, in his Sanskrit textbook, that this is partly due to regional variations in India and partly due to Anglisation. Other authors point out that words like jñāna were taken into modern languages as loan words and that this is the source of regional variants.

In the yellow and orange coloured regions, and in Panjab, the pronunciation is /gya/. In the green and red regions and Sindh, it is /dnya/, although in Gujarat it is /gnya/. South India has a mixture. English speakers often opt for /nya/. We can organise these variants: 

IAST
jña	→ g ña (gnya)	→ gya
	→ d ña (dnya)	→ dya
	→ ña (nya)
IPA
[ɟɲɐ]	→ [g nʲɐ]	→ [gʲɐ]
	→ [d nʲɐ]	→ [dʲɐ]
	→ [nʲɐ]

The change from [ɟ] → [g] is a move back in the mouth towards the velar articulation while the change from [ɟ] → [d] is a move forward towards the alveolar. These are clearly different processes and one could not lead to the other. However, with the shift in the initial consonant, the nasal changes from the plain palatal nasal [ɲ] to the voiced palatised nasal [nʲ]. This is why we see it written as "nya". In fact, what happens is the tongue makes contact at the alveolar ridge and rolls back along the palate. In Sanskrit we'd say it was ī blending into a (or a backwards diphthong ai). Phonetically we'd write the vowel change as [ɨ] → [ɐ]. In a sense, then, [nʲɐ]  sounds like [ɲɨɐ].

We can infer that gya [gʲɐ] could be a simplified version of gnya [g nʲɐ]. Thus the Gujaratis who pronounce jñā as gnya [g nʲɐ] are probably conservative with respect to the northern and eastern Indians who have reduced this to gya. Since English speakers have difficulty with [ɟɲə] they simply reduced it to [nʲə]. Pāli generally reduced conjunct consonants to geminate consonants thus jña becomes ñña. But the ñ is pronounced by English speakers as a palatalised nasal not as patal, i.e. as [nʲə] rather than [ɲə], hence names like Nyanaponika (Ñāṇapoṇika) and Nyanamoli (Ñāṇamoḷi).

Since old Persian contains the regular change [dʒ] →  [d] we might suspect from Persian influence on the modern regional pronunciation in parts of India most influenced by the later Persians. However, this does not quite add up since Persian speaking Mughals also ruled in Delhi where /gya/ is standard.

One thing to keep in mind is that in Latin -gn- was pronounced [ŋn], i.e. they did not pronounced gnōsis as /g-nosis/ but as /no-sis/ (like Skt ṅosis). However, the sandhi rules of Sanskrit tend to rule out this possibility.

Sandhi

All spoken languages exhibit changes in pronunciation as words and sounds run together. Some of the languages that use a phonetic writing system notate them and some don't. Sanskrit does and there are a set of rules called sandhi (joining). In classical Sanskrit the rules are compulsory and applied uniformly, but in vernacular languages they became more optional. Madhav Deshpande pointed out to me (in an email) that certain sandhi involving j make it clear that it was heard as [ɟ] even in the conjunct jñ.

The Sanskrit Heart Sutra gives us an excellent example. Towards the end it says "therefore it should be known", in Sanskrit without sandhi: tasmāt jñātavyam; with sandhi tasmāj jñātavyam. The rule is t followed by t becomes j. Similarly in the Mahābharata (179.1) we see  tasmāj jīvo nirarthakaḥ "Therefore this life is useless", and in the Bhagavatā Purāṇa (11.19.5) we see tasmāj jñānena "therefore by this knowledge",

The rules are very specific and became part of the language long before they were codified by Pāṇini ca 4th Century BCE. This tells us that the j in jña was being heard as [ɟ] in Sanskrit when the rule was made or else an exception would be noted.

Chinese

Another potential source of information is Chinese transliterations. Prajñā, for example, was a word that was generally transliterated rather than translated. The transliteration was established early on, by Lokakṣema ca 179 CE, as 般若; however, this was based on Gāndhārī texts. In modern Buddhist circles this is pronounced bōrě although the standard Pinyin transliteration is bānruò. The Middle Chinese reconstruction is complex. There are multiple choices and multiple encoding systems.

• 般 has been reconstructed as /puɑn/ or /pʷɑn/ or /pan/ or /pɐn/.
• 若 has been reconstructed as /ȵɨɐk̚/. The symbol ȵ is mainly used in Sinological circles and represents the alveolo-palatal nasal, i.e. [nʲ].

The Gāndhāri spelling of prajñā was praṃña or praña (c.f. Pāḷi paññā) And thus we can see that the standard Chinese transliteration presents a Prakrit pronunciation without the conjunct. Unfortunately, it was never updated to reflect the Sanskrit pronunciation when Sanskrit texts started to appear. And thus we cannot use the Chinese to reconstruct the received pronunciation at any later period.

Writing

Since Sanskrit is usually pronounced as written we can also look at historical writing to see what it tells us. In modern Devanāgarī the akṣaras for ja and ña combine to form a special sign for the conjunct jña, i.e. ज् + ञ > ज्ञ. Hiralal Rasikdas Kapadia (1936: 289), an expert on Jaina literature, points out that there are two types of Devanāgarī and the Jaina variant as another way of writing jña. In the Jaina variant we get more of a sense of the conjunct. This becomes clearer in old Indian scripts.

Older scripts make a clear distinction between the j and the ña when writing jñā. A good point of reference for us is the Hōryūji manuscript which probably dates from the 9th Century. This manuscript contains a syllabary. The image on the right shows the palatals: ca cha ja jha ña.

We can also isolate the akṣara jñā from this manuscript and see that it has two quite distinct ligatures and is thus written as a conjunction of j and ñā. Note that when written as a ligature the shape of ña is quite different. Also, the scribe has adapted the the basic ja shape in order to indicate the long vowel ā. I will provide some handwritten examples that make the process clearer.

In the collection of akṣaras below I try to show the evolution of ja from Brahmī which gives us two distinct forms of ja, one of which informs the Devanāgarī script and one of which is preserved in Japanese Siddham.

Left to right: Brahmī, Gupta, Siddham (Hōryūji ms.);
upper: Siddham and alternate; jñā (Hōryūji);
lower: Nāgarī, Devanāgarī.

In the formal Siddham script of medieval Japan, the ñā ligature is often written vertically, but this makes for a very tall character. The Hōryūji scribe turns the ligature through 90°. Here are some original and copied examples of jña and jñā. In modern Siddham we often see jña written for jñā.

Some of this is superfluous detail, but it makes the point that until the invention of the (non-Jaina) Devanāgarī script jñā was written as a standard conjunct made of ja and ña. Had the Sanskrit pronunciation been meant to be something else, then the scribes would have used another akṣara.

Note: 18 March 2020. The Creation and Spread of Scripts in Ancient India. Harry Falk

Conclusion

On the grounds of historical phonology, sandhi, and graphology we can say that jña was intended to be pronounced [ɟɲɐ]. The sandhi rule for tasmāt jñātavyam makes it clear that at some point this was the pronunciation and there was no parallel of the Latin pronunciation of [ŋn] for gn. The regional pronunciations are logical developments from an original [ɟɲɐ] and influenced by the use of Sanskrit loan words. However, we don't really know why the pronunciation changed, only that it did. 

For Indian pandits and academic Sanskritists alike, the pronunciation they learned from their teachers will likely be the way they keep pronouncing it. Furthermore, the sound combination [ɟɲɐ] is difficult for English-speakers and they will inevitably tend to drop the [ɟ] and end up with [ɲɐ] or [nʲɐ]. So probably few will follow me in adopting [ɟɲɐ]. And thus all this is for nought. But I find it fascinating and it has kept me amused for a week. It was nice to get my calligraphy pens out again after a long break. In these difficult times, that's not a bad thing, eh? 

~~oOo~~

Bibliography

Burrow, T. (1973). The Sanskrit Language (3rd edition). London: Faber and Faber.

Kapadia, H. R. (1936). 'A Note on Kṣa and Jña.' Annals of the Bhandarkar Oriental Research Institute. 17(3), 1935-36: 289-296.