Matrika and Malini: why one Sanskrit alphabet is out of order

In 1997, at a seminar in All Souls’ College, Oxford, a scholar named Somadeva Vasudeva argued that an old list of Sanskrit sounds was a picture of a woman.

He was then at work on the critical edition of parts of the Malinivijayottaratantra that he finished at Oxford in 2000. The paper took ten years to reach print. When it appeared in 2007, in a memorial volume for Helene Brunner, it announced itself as the first of a triptych. We could not find the second and third parts.

The list he was talking about is an ordering of the Sanskrit alphabet. Not the ordering a child learns. A different one, taught in Shaiva scripture and used in ritual, which begins with the sound NA and ends with the sound PHA, and in between puts the vowels and the consonants through each other in a sequence that looks, on the page, like a shuffled deck. The tradition names it out of its own first and last letters: the nadiphantakrama, the NA-to-PHA order.

Abhinavagupta, writing in Kashmir around the year 1000, called that sequence the principal one.

He did not say why it was in that order. Neither did the scripture he was quoting. His commentator Jayaratha came closest, and what he supplied was a restatement: she begins with NA and ends with PHA, he wrote, because her seeds and her wombs are mixed up.

Be careful what follows from that, because there is a version of it that flatters the modern scholar at the tradition’s expense.

The tradition does say the sequence is mixed up. Jayaratha says it in the sentence above. What the tradition does not say is that it is akrama, without order, and the distinction is technical and sharp: akrama in this system is a word reserved for the highest reality, Shiva’s own autonomy, so a sequence of phonemes is not the sort of thing that is permitted to be disordered. Vasudeva’s own abstract grants that the order is allegedly disordered and undertakes to show the disarray is only apparent. Nobody in the tradition thought the letters fell where they fell at random; nobody in the tradition said why they fell there either.

The flat no-order reading is modern, and Vasudeva names a bearer of it. Jaideva Singh, whose translations put a great deal of Kashmir Shaivism into English, describes the Malini order in passing as arranged in an irregular way, the vowels and consonants mixed, no serial order observed. The charge against him is not that he argued the wrong thing. It is that earlier scholars “have presumably not reflected on the matter.” An unexamined default, held by people who had no particular reason to examine it.

There is something worth noticing about where Vasudeva found that sentence. It comes from Singh’s own running exposition of Abhinavagupta’s commentary on the Paratrishika, which is to say from an exposition of one of the very texts Vasudeva then quotes to prove the order is real. Bettina Baumer quotes the same sentence in her introduction to the same volume, so it is possible Vasudeva took it from her rather than from Singh’s page.

The alphabet that comes apart

Start with what is not in dispute.

The Sanskrit syllabary has a canonical order, and it is not arbitrary. The vowels come first, in pairs of short and long, and they are conventionally counted as sixteen because the run closes with two sounds that are not vowels at all: the nasal hum written as a dot above the line, and the breathed release written as two dots after it. Then the stops, in five rows of five, arranged by where in the mouth they are made and whether they are voiced and aspirated. Then the four semivowels, the three sibilants, the aspirate, and last KṢA, a cluster of two consonants that is counted as a letter in its own right. Sixteen and thirty-four. It is a phonetician’s table, and it was a phonetician’s table two thousand years before the International Phonetic Alphabet. Fifty sounds, filed by the anatomy of the mouth.

Individual phonemes appear in capitals through what follows, carrying the diacritics the Sanskrit distinctions require. English spelling runs together sounds this material keeps strictly apart, and the difference matters here more than usual, because these letters are about to be assigned to separate parts of a body. ṬA is not TA. ŚA, ṢA and SA are three different letters.

In tantric Shaivism that order is a goddess. She is called matrika, and the word simply means alphabet: it is the ordinary Sanskrit term for the set of letters, the way abecedary once was in English. Deified, she is the source of every word, every scripture, every mantra. One layer of the Nishvasa corpus, the oldest surviving body of this literature, says there is no mantra beyond her.

And then there is a second sequence of the same fifty sounds, out of order, also a goddess, called malini, the garlanded one.

The contrast everyone reaches for is matrika against malini, the regular against the scrambled. It is a clean opposition and it is not quite the tradition’s own.

In the scriptures of the Trika, the Kashmirian Shaiva system named for the triads it organises everything into, the ordinary order’s presiding deity is not matrika at all. He is male, and he is called Shabdarashi, the Mass of Sound. Vasudeva notes this at the head of his chapter and then, across thirty-four pages, uses matrika in the other sense exactly once, in a footnote. Judit Torzsok, tracing the earliest layers, finds Shabdarashi arriving abruptly in the Siddhayogeshvarimata and being very poorly looked after: he turns up in what appears to be a later appendix, is mentioned once elsewhere, and his version of the alphabet code is hardly used. Her verdict is that he is “something of an alien here.”

So the pair is not originally a pair. It is three deities, unevenly distributed across texts that do not agree, and one of them is a man.

Both of them were called Malini

The sharper fact is terminological, and it undoes the headline.

In the Malinivijayottara, the root scripture of the Trika and the text Abhinavagupta claims to be expounding, the Goddess asks a question in chapter four. She has already been taught how the levels of reality are disposed in one body. Now she asks about the other. The two bodies she names are abhinnamalini, the Malini whose womb is undivided, and bhinnayonimalini, the Malini whose womb is split.

Both orderings, in that text’s own idiom, are Malini. What distinguishes them is whether the vowels and the consonants are kept apart or run through each other. The ordinary alphabet keeps them apart: all sixteen vowels, then all thirty-four consonants. The scrambled one interleaves them. Undivided womb, split womb.

The clean two-name opposition, matrika here and malini there, is later tidying. And the tidier is Abhinavagupta.

He is good at it. In the Tantraloka he makes the three deities into one process: the Mass of Sound, considered as a single act of awareness, is Bhairava; that same power, touched by the shadow of what is to be known, is Matrika; and she, through samghatta, the violent rubbing together with the Mass of Sound, becomes Malini, her womb split. Three names, one thing, in sequence. It is elegant and it is an argument, not a report.

The pun is better than the system. Bhinnayoni means the womb is split. It also means the consonants are out of sequence. Jayaratha spells the double sense out: she begins with NA and ends with PHA, he says, because her seeds and her wombs are mixed up, the seeds being the vowels and the wombs the consonants. The scrambling of the alphabet and the opening of a body are the same word.

You can watch the substitution happen inside a single phrase of his. In the commentary on the Paratrishika he sets bhinnayonimalinibhattarika, the Lady Malini of the split womb, against abhinnayonimatrika, the undivided-womb Matrika.

The modifiers are the scripture’s. Vasudeva, who edited the text, reports its own pair of names as the split-womb Malini and the undivided-womb Malini. What changes in Abhinavagupta’s phrase is the noun. Where the scripture has two Malinis distinguished by a modifier, he has a Malini and a Matrika, and the distinction has moved from the modifier into the names themselves.

That is a small thing to do to a word, and it is the whole opposition every later account of this material inherits. Including this one. There is no way to write about the subject in English without using matrika and malini as a pair, because that is how a thousand years of commentary, and most modern scholarship, have named it. What can be done is to say once, plainly, that the pair is the exegetes’ and not the scriptures’, and to remember it at the points where it matters.

Where the goddess of the alphabet came from

The alphabet goddess is not a Shaiva given. She is a Shaiva development, and we can watch her arrive.

The evidence is the Nishvasa corpus, the earliest surviving body of this literature, which survives in layers that can be put in order. Torzsok read them in sequence and published what she found as a conference paper in Pondicherry in 2009.

In the oldest layer, the Mulasutra, there is no Matrika. There is Vagishvari, the goddess of Speech, twice, and in her the initiate is ritually reborn in the liberating initiation. Speech is already a goddess. The alphabet is not.

She is missing from outside the corpus too, in a text Torzsok brings in as a parallel rather than a stratum. The Vinashikhatantra belongs to a different current, the Vama, and it prescribes what later texts call the alphabet grid while naming no goddess at all. It locates the power of the syllables in Shiva: the body of the god is the multitude of all seed-syllables, presided over as the stars are by the pole star, and the world up to the world of Brahma is made of the letters from A to KṢA. Torzsok’s comment on the silence is properly cautious. Had the text known of her, it “could have or perhaps would have not omitted her name in this context.” She reads the Rauravasutrasamgraha the same way, as a text that gives mantric power to Shiva or to Shiva’s power without clearly personifying it, and calls it a transition between the two ideas.

Back inside the corpus, in the Uttarasutra, she appears, as a vidya, a female mantra: the source of everything made of words, of mantras, of treatises, consisting of eight groups of letters. But she has no gender of her own yet. The text calls her matrikashiva, a compound, Matrika-Shiva. And it makes a claim this literature never abandons: there is no liberation like initiation, and no mantra beyond Matrika.

In the Nayasutra the body arrives. The practitioner imitates each letter with his own body, and both Matrika and each single letter, the letter A in particular, are said to contain all the levels of reality. He who knows this form of Matrika in his body, and likewise in the letter A, is released.

In the Guhyasutra she is unambiguously a goddess, because Shiva addresses her as one. Everything is produced from my seeds, he says, and is born from your womb. And what is produced is specifically verbal: the Vedas, the systems of thought divine and semi-divine and human, and languages. Sanskrit, Prakrit, and even Apabhramsha, which is to say the ancestors of the modern vernaculars. The alphabet goddess gives birth to grammar and also to slang.

Then the Svacchandatantra sorts her into eight groups of letters and maps the eight Mother goddesses onto them. That much is secure; the lines that actually name the Mothers are a later insertion, and the argument about them is in the disagreements below.

And in the Brahmayamala, the earliest surviving scripture of the yogini cults, there is a passage that is the best image in this entire literature. Torzsok transcribed it from a Nepalese manuscript, in a chapter on procedure, describing the level of maya. The goddess there has many cells and resembles a mass of diluted yellow orpiment. She has the form of a honeycomb. She looks like the petals of a red lotus, and pervades whatever is made of speech with her great strength. She is established in a bee-dwelling and is constantly agitated by the Lord, and by that agitation the triple world is always created. In the place of innumerable cells, in thousands of ten millions of them, other forms are established by their own nature. She has the form of fiery energy everywhere, like a firefly in trees.

A honeycomb of cells, shaken until a world comes out of it, glowing like fireflies in a tree. Torzsok notes in passing that kshobha, the word for the shaking, and its derivatives in this text often appear to mean bringing to orgasm.

One correction has to travel with that passage wherever it goes, and it usually does not. The goddess described there is called Malini, and she is not the Trika’s Malini.

She is a form of maya, the power by which the world is spread out and taken for real. Her phonemes number forty-nine rather than fifty. And the NA-to-PHA sequence, Torzsok says, does not seem to be mentioned in the Brahmayamala anywhere, so that the goddess described there “cannot be considered identical with the well-known alphabet goddess of the Trika, but she is certainly an alphabet goddess.”

Which forty-nine she is remains open, and the words that would settle it are the damaged ones. Hatley’s conjecture would make her set the ordinary run ending in KṢA; Sanderson preferred to keep what the manuscript has, which would give something nearer “in small compartments” and say nothing about the range. The bee-dwelling itself is Sanderson’s conjecture, replacing what the manuscript has. So the passage above is a translation of a text that is in places a reconstruction, and the most memorable phrase in it is one of the reconstructed parts.

A honeycomb. An alphabet goddess named Malini. A different one.

And the word goes on being borrowed. In a Shaiva Siddhanta work on the nature of sound, Ramakantha, whom Williams places around 950 to 1000, lists the names under which one cosmological principle appears across the scriptures; malini is in the list, between Sumangala and Mahamaya. There it is not an alphabet goddess at all. It means the supreme resonance out of which all sound is made, set above maya in that school’s map of the real. Right word, different thing again. The word was common property and the referent was not.

Which is itself the finding. There was more than one alphabet goddess, and more than one Malini, before anybody tidied up.

The Siddhayogeshvarimata has both goddesses present, praises Matrika in the usual terms, and quietly does its real work with Malini, whose limbs are letters and whose code encodes the text’s mantras throughout. Her name is rarely spoken there. Usually she is just “the goddess”, or “the power of Rudra”, or “she who is made of all letters.”

Torzsok’s own summary is that the power of letters was attributed to Shiva first, became increasingly personified, and only then separated into deities. She also warns, twice, that this is a reconstruction from what happens to survive, that it is difficult to conclude anything from the absence of a thing, and that “the above way of seeing how Matrika’s cult evolved may be very distorting.” Keep the hedge. The strata are real; the arrow drawn through them is a reading.

One correction worth making here, because it is everywhere in the secondary literature. The Seven or Eight Mothers, the warrior goddesses of the Puranas and the temple panels, are matr, mothers, without the suffix. The alphabet goddess is almost always matrika, with it. Calling the Mothers saptamatrikas conflates two things the sources distinguish. Torzsok insists on this, and the tradition backs her: Narayanakantha glosses the suffix with iva, “like” a mother, and Abhinavagupta builds an entire etymology on it.

Though the texts then go and merge them anyway. The Svacchanda divides Matrika into the eight Mothers; the Siddhayogeshvarimata makes Shabdarashi’s body consist of eight Mothers standing for eight sections of the alphabet. The concepts are distinct in vocabulary and interpenetrating in ritual, which is a more interesting answer than either one or two.

The settlement, when it comes, is a one-word aphorism and a commentary on it. The Shiva Sutras, the terse Kashmirian aphorisms on which the whole later tradition hangs, give as the seventh aphorism of the second section a single compound: matrikachakrasambodha, the full awakening to the wheel of the Matrika. That is the entire sutra. Kshemaraja reads it as continuous with the aphorism before it, which says that the guru is the means, so that the sense is: from that guru, when he is pleased, the full awakening to the wheel of the Matrika comes to the disciple. Then he says he will set it out by the method taught in the Paratrishika and the rest, and sends the reader to Abhinavagupta’s commentary on that text for the details.

What follows in his own pages is not a definition of the alphabet. It is a cosmology read off it. A is the first phase of I-awareness; the long Ā is that same awareness expanding; I and the long Ī are the stage of will and its lordship; U is the opening, with knowledge; the four vocalic liquids are will coloured by the merely sought and so are called the four neuter seeds, incapable of generating anything further; E is the triangle-seed, AI the six-cornered, AU the trident. Then the dot, then the emission. Then outwardly the five rows of five, unfolding the levels of reality down to the person. The semivowels are called “those standing within” because they abide as the sheaths that bind a soul. The three sibilants and the aspirate together are the blazing ones, because they blaze up as difference dissolves into non-difference.

That is where the claim “the alphabet is the universe” is actually argued, rather than merely asserted. And note which alphabet it is argued on. The ordinary one.

The standard modern account of this whole body of doctrine is Andre Padoux’s Vac, which gave English the phrase “phonematic emanation” and documents Abhinavagupta’s version of it across seventy pages. Ben Williams, surveying the field, calls it a truly pioneering and far-ranging monograph and then names what it does not do: it achieves a global perspective “but without tracing shifts in intellectual and conceptual histories, particularly between the milieux of scriptural redactors and later expositors.”

That gap is the subject of this article. Vac sets out the mature system magnificently; the question of what the scriptures were doing before the exegetes arrived to harmonise them is the one Torzsok and Vasudeva went after. Williams puts the harmonising on a date and gives it names: Shrikantha in the early tenth century, Ramakantha around 950 to 1000, Kshemaraja, and Abhinavagupta. And for one branch of it he quotes Dominic Goodall on a finding that should unsettle anyone reading an exegete as a report of scripture: “there may in fact be no extant scriptural source that presents the emanation and nature of sound in the way that the writers of the mature Kashmirian Shaiva Siddhanta do.”

That is said of the Siddhanta commentators and not of the Shakta texts this article is mostly about, and Williams is careful to keep the two apart. But it establishes the shape of the problem. A mature doctrinal system can turn out to have no scripture behind it at all.

The alphabet in the future tense

How many letters is she? Fifty, usually: sixteen vowels and thirty-four consonants, and the Svayambhuvasutrasamgraha says exactly that. Forty-nine in the Brahmayamala, by a reckoning its damaged text does not let us pin down.

The instability is not sloppiness. Torzsok shows that the Siddhayogeshvarimata adds KṢA as a separate final consonant precisely so that eight Mothers can be fitted to the consonants alone, because that text wants the vowels male and the consonants female and therefore cannot give a Mother to the vowels. The count of the alphabet is downstream of how many goddesses you need to house in it. A figure of fifty-one circulates in later reckonings; nothing we read attests it, and we are not going to print it as though something did.

What the exegetes do with the names is the better material, and it is where this tradition’s relationship with the grammarians actually lives.

And there is a third ordering of the sounds of Sanskrit, which an article asking why one alphabet is out of order has no business leaving out.

Panini opens his grammar with fourteen aphorisms that list the sounds of Sanskrit in an order of his own, grouped so that a single abbreviation can pick out any class he needs. It is a scramble in the service of economy. Tradition calls those aphorisms the Maheshvarasutras and holds that Shiva revealed them, and at least one later work, the Nandikeshvarakashika with its commentary, gives them a fully Shaiva metaphysical reading. So there are three orderings in play, all three in some sense attributed to Shiva, each arranged on a principle the other two have no use for: the mouth, the body, and grammatical shorthand. Whether they are three arrangements of one identical set we did not establish, and it is unlikely, since a grammarian’s inventory and a tantric one are drawn up for different purposes. So take the comparison as three ways of ordering the sounds of the language rather than three shuffles of one pack of fifty cards.

What has anyone made of that juxtaposition? Nothing we could find, and we are not going to imply a literature we did not locate. On the wider question of the tantric exegetes and the grammarians there is at least a place to start, Raffaele Torella’s study of Bhartrihari in Kashmirian Shaiva non-dualism, in a volume given over entirely to the linguistic traditions of Kashmir; we did not get hold of it. On the three orderings standing side by side we found no scholarship at all, and the contrast is noted here as a contrast and nothing more.

Start with the name itself, where there is a real surprise waiting. Everyone reads matrika as mother; the tradition reads it that way; and the doctrine that she is the womb of all speech leans on it. Judit Torzsok has raised the possibility that the alphabet sense does not come from matr, mother, at all, but from matra, the prosodic mora, the unit by which the length of a vowel is measured. That would make her a goddess of measured sound, and the whole maternal theology a folk etymology that hardened into doctrine. Two cautions. The suggestion reaches us at second hand, through Ben Williams’s summary, where it is reported as speculation; and we could not get the page in her own text. It is a possibility worth knowing and not a finding.

What is documented, and it is better than a quarrel, is tantric authors running the grammarians’ own machinery as theology, at full speed, with no sense of trespass.

Abhinavagupta derives matrika from the root ma, to measure or to experience, with the agent suffix -tr read as forming a periphrastic future, and the suffix -ka read as the suffix of comparison. The result is that Matrika is she who is like the one who will be the subject of experience. She is still inside the Mass of Sound, unmanifest. Only when she becomes manifest is she Malini.

Which is to say: matrika is the alphabet in the future tense. That is a Paninian derivation deployed to generate a metaphysical difference between two alphabets, and it works.

Jayaratha offers three derivations of malini and does not choose between them. From a verbal root al- that appears to be his own invention, taken with the prohibitive particle ma, in the sense of opposition, so that she is the one who holds the consciousness of it. From malate, to hold or to possess, so that “she holds the world in her own form”, and for that sense he goes to the Dhatupatha, the grammarians’ own catalogue of Sanskrit verbal roots, which is where a grammarian would go to establish that a root means what he says it means. And from the denominative malayati, she interiorises the whole universe.

And in the Tantraloka’s third chapter Abhinavagupta takes the grammarians’ label for the semivowels, antahstha, “standing within”, and derives it afresh as a cosmological fact: that is exactly why this tetrad is called antahstha, because of being internally contained within desire and the rest. The same move is made on vyanjana, consonant, and sparsha, contact, which he gives as names for congealed emission.

The emission, thus released, having arrived at congealment, is termed breath, consonant, and contact.

That is the doctrine in one line. A consonant is not a symbol of divine emission. It is divine emission, gone solid. The phonetician’s vocabulary is accepted entirely and claimed never to have been merely phonetics.

What the rite actually does

The operation both alphabets exist for is nyasa, installation: putting the phonemes onto and into the body. Like bhuta shuddhi, the purification of the elements of the body, it is something a practitioner does before worship rather than instead of it.

Abhinavagupta sets the two rites side by side in the chapter of the Tantraloka on daily worship. The frame verse does a lot of work in one line: one should perform the installation in a general way, for the accomplishment of the external worship, the matrika, or the malini, or both, in sequence or out of sequence. Three things follow immediately. It is preliminary to external worship, not a climax. The choice is open. And “or both” settles that they are not rivals.

But the choice is not evenly weighted, and this is where the two rites come apart. Where no special prescription governs, the Malinivijayottara says to install the split-womb Malini, and it says what for: for the sake of a shakta body, a body made of power. Abhinavagupta quotes that line twice, versifies it into his own chapter, and adds that this scrambled sequence is the principal one even though the intermixing of womb-phonemes and seed-phonemes obtains in many forms. So Malini is the default and the ordinary alphabet is the alternative, which is the reverse of what the names suggest.

The two body-maps are built differently, and the difference is the point.

Matrika-nyasa proceeds by groups. The sixteen vowels go on the head and face: forehead and mouth, then in pairs on the eyes, ears, nose, cheeks, teeth and lips, then the topknot and the tongue. Two whole rows of stops go on the right and left shoulders, arms, hands, fingers and nails. Two more from the hips and thighs downward. The pa row on the flanks, back, belly and heart. And the last nine on the bodily constituents: skin, blood, flesh, sinew, bone, fat, semen and what they head.

Note where that ends. Not on the surface. Matrika-nyasa finishes inside the tissue. Whole rows of the alphabet mapped onto whole sets of limbs, the syllabary’s own order carrying the body’s order along with it.

Malini-nyasa is a single running thread. One phoneme, one place, from the crown downward, in the scrambled order, and it assigns things the other map has no slot for: the trident and the skull-bowl in her hands, the breasts, the milk, the life-breath. The body is read off the sequence, and the sequence is the body.

The earliest attested form of all this is neither, and it is stranger than both. Dominic Goodall, editing Nishvasa Guhyasutra 9, found that when the Goddess asks how power and liberation come about through the matrika, Shiva’s answer is not a technique. It is a costume. The practitioner is to dress half as a woman and half as a man: feminine tresses on one side and matted locks on the other, a forehead mark on one side and a third eye on the other, a ring in one ear and a pendant in the other, a trident in the right hand and a breast on the left, a woman’s anklet left and a man’s anklet right. In the observance of Ardhanarishvara, the Lord who is half woman. Then a month of alms and controlled senses and barley-gruel; low attainments after two months, middling after six, high after a year.

Goodall’s explanation is the valuable part. Dressing as Ardhanarishvara suits the matrika because the matrika is made of feminine vowels and masculine consonants, and these are applied respectively to the left and right halves of the practitioner’s body before worship, in a preliminary rite that prefigures what came to be called sakalikarana, the making of the body complete by installing mantras in its parts.

Hold that polarity loosely. Which sex the vowels are is one of the live disagreements in the current scholarship, and the section below on what the scholars cannot agree about sets out why. Goodall’s reading of the costume depends on his answer.

So the first version of letters-on-a-body is not a fifty-point map. It is a body cut down the middle, vowels one side, consonants the other, and the way you perform it is by getting dressed. The elaborate maps are a refinement of a cruder and far more arresting gesture.

As for what nyasa is for, Abhinavagupta answers that too, and the answer is maintenance. When the rite, or the potency, or the mantra’s own form has lapsed, mantras installed by this are refilled by a repeated installation and become fruit-giving again.

Mantras leak. Nyasa tops them up. Shaman Hatley puts the general point well: this process of mantra-installation is in fact the very template of ritual, inner and outer, with elaborate and abbreviated forms. It is not an initiation, not a diksha. It is the thing you do before the thing, every time, for years.

Nor, on the evidence of this chapter, is either rite restricted by grade of initiation. Both sit in the chapter of ordinary worship. Where initiation does bite, on Shaman Hatley’s account of the Brahmayamala, is on something else: the practitioner’s personal pantheon is fixed at initiation, and that is what the installed mantras belong to. The choice of alphabet is not what the grade governs.

Read backwards

One thing an account of this material gets wrong if it stops at NA to PHA.

The sequence also runs the other way, and the other way is doctrinally specific. Abhinavagupta describes uttering it from PHA back to NA, and says that in doing so one withdraws the circle of the cosmic paths, the shadadhva, after which consciousness is full, and how then is she to be described in words. Forwards is emission. Backwards is reabsorption. PHA to NA is not a rival ordering; it is the same ordering run in reverse, for the opposite purpose.

The Malinivijayottara confirms it structurally. When it maps the levels of reality onto the scrambled alphabet in chapter four, it starts at the far end: in PHA the earth-principle is taught, then DA and the rest in order, and the sixteen from GA to NA are in the Shiva-principle. Earth at the last phoneme, Shiva at the first. The alphabet read forwards is the world coming out; read backwards it is the ladder of the real.

Abhinavagupta also concedes, in the same passage where he declares this sequence principal, that other scrambles exist. Although the intermixing of womb-phonemes and seed-phonemes obtains in many ways, he writes, nevertheless this one is proclaimed the principal one. He is not claiming there is one scrambled alphabet. He is making a choice among several and telling us he is making it.

And the cleanest statement of the whole relationship is one sentence of his commentary on the Paratrishika, where he says that the same phonemes hold two different positions among the levels of reality depending on which ordering you read them under, and quotes the Malinivijayottara’s verse to prove it. The two alphabets are not alternatives. They are two readings of one set of fifty sounds.

He puts it more sharply elsewhere in the same work. The Heart of Bhairava, he writes, is born with reference to the Malini from NA, but in reality from A.

Same reality. Different first letter.

Which is elegant, and it still does not say why NA. A sequence can be run forwards and backwards and mapped onto the whole of reality and remain, as an arrangement of letters, perfectly unexplained. Before anyone asks why these fifty sounds are in this order, there is a prior question, and it is the practical one: what was the order for?

Why anyone would want a scrambled alphabet

Here is the functional answer, and it is the part most accounts leave out.

Once a body has been lettered, a mantra can be dictated as anatomy.

The Malinivijayottara does exactly this. Having given the body-map in chapter three, it says: from this, hear how the vidyas and the mantras are to be extracted. And then it spells out the mantras of the three goddesses Para, Parapara and Apara by naming body parts. The right shank with a dot on it. The living. The left shank. A reader who has the map reads off the syllables. A reader who does not read it reads a list of limbs.

This is mantroddhara, mantra-extraction, a real genre with its own conventions, in which the potent form of a mantra is specified indirectly rather than set down plainly. The lettered body is one of the devices that makes it possible. The Malinivijayottara gives the map, then uses the map, and the second half of chapter three is unreadable without the first.

It is tempting to call that a cipher, and the temptation should be resisted, because the article’s own later evidence will not support it. A cipher needs keys that differ. This one does not: as the next section will show, the letter-to-limb assignment barely moves between the texts that transmit it, and the map is written out in the revealed scripture itself. Anyone who owned the text owned the key. Whatever the lettered body was protecting the mantras from, it was not readers of the Malinivijayottara.

What it does achieve is something less dramatic and more interesting. A mantra specified as anatomy is a mantra that cannot be uttered by accident off the page, that has to be reassembled deliberately, part by part, by someone who has already installed those parts on himself. The extraction is not a lock. It is a procedure, and performing it is the point. No scholar we read says this in so many words, so take it as our reading of the texts rather than a finding.

Either way something was built on the lettered body, and the question of why the letters sit where they sit is not answered by saying that mantras were extracted from them. A code only needs to be settled. This one is settled in a particular arrangement rather than any other, and the arrangement is what wants explaining.

Why these letters on these limbs?

The answer was a drawing

Vasudeva began by refusing to accept that the order was no order.

His first move was doctrinal rather than philological, and it is a nice piece of reasoning. In this tradition, he points out, akrama, absence of order, is a technical term reserved for the highest reality. Abhinavagupta uses it of Shiva’s own autonomy and of nothing less. A sub-divine sequence of phonemes therefore cannot be akrama, on the tradition’s own terms. Whatever the sequence is, it is ordered, because in this system only God is not.

His second move was to count. He surveyed thirteen passages teaching the sequence and tabulated eleven distinct versions of it, from the Siddhayogeshvarimata, the Malinivijayottara, the Agnikaryapaddhati (a Kashmirian ritual manual surviving in manuscripts at Gottingen and Paris), the Tantrasadbhava, the Kubjikamata, the Trishirobhairava as Jayaratha quotes it, two recensions of the Shatsahasrasamhita, the Shrimatottara, the Kularatnoddyota, and a badly corrupted passage of the Agnipurana from which only twenty-nine phonemes can be recovered. Only five had been published before.

They disagree a great deal, and it is worth being precise about how much, because the usual summary of this argument is tidier than the evidence.

The sequences diverge badly. The oldest of them, in the Siddhayogeshvarimata, is in Vasudeva’s judgement “the most eclectic of any work consulted” and only remotely related to the rest, and he goes further: of that version, “the very name nadiphanta is inappropriate since the sequence neither begins with na nor ends with pha.” The earliest witness to the NA-to-PHA order is not in NA-to-PHA order. One of the eleven, the Agnipurana passage, is so corrupt that twenty-one of the fifty assignments cannot be read at all.

And across all of that, two things hold. The first is a pantheon: every one of the fifty phonemes has a goddess presiding over it, variously called a Shakti or a Kala or a Yogini, and which goddess goes with which phoneme barely shifts between texts. The second is the one Vasudeva builds on. His own statement of it is the hinge of the entire argument:

“Even more striking than the fidelity with which the phoneme-Sakti assignation was retained, is the flawless preservation, despite the frequent divergence of sequence, of the association of a corresponding phoneme and body-part. This shows beyond doubt that this is the one key feature essentially defining the nadiphantakrama in the pre-exegetical tradition. Astonishingly, there occur no variations at all in any of the texts consulted.”

Read that carefully, because the two halves are doing different jobs. “Despite the frequent divergence of sequence” concedes the mess. What never varies is not the order of the letters. It is which letter goes where on the body. A transmission spanning several centuries, redactors reshuffling the sequence freely, and nobody ever moving a letter off its limb.

Even there the fit is not perfectly flush, and his own table shows it. Several locations carry text-specific glosses: NA is the topknot, or in one text the tip of the topknot; the long Ī is the nose, or in another the breath that passes through it; SA is the soul, or the soul in the heart, or the supreme soul. These are refinements of the same place rather than relocations, which is why he can say what he says. They are also the reason a careful reader of his table will not describe the mapping as identical, only as unmoved.

Which inverts the problem all the same. The order is not the stable thing. The map is, and an order is what you get when you read the map in some particular direction. Vasudeva’s third move was to identify the direction: head to toe, which is the direction of the installation rite.

Two things should be said about that step, and the article that reports it owes them to the reader. The one external witness to convention points the other way: the Kashmirian commentator Vallabhadeva records that goddesses are conventionally described upward from the toes and only humans downward from the hair, so a descending goddess is the exception, and Vasudeva accounts for the exception by having her manifest in the order of creation rather than of reabsorption. And he then uses the descending order as a tool on the texts, settling a disputed word as a head-chaplet rather than a necklace because of where it falls, and diagnosing a displaced letter in the Tantrasadbhava as a redactor’s error. Which means at least one divergence from the reconstruction has been reclassified by the reconstruction as a mistake.

Here is the consequence, and it is the thing to hold on to. Begin at the topknot, descend to the feet, write the letters down in the order you touch them, and you do get the sequence. In the Malinivijayottara. That is why he calls its version the most systematic of the eleven and reads its four and a half verses as an anatomy rather than a list. In the Siddhayogeshvarimata you do not, and he says so himself.

So what the drawing explains, strictly, is one witness out of eleven: the cleanest one, the one Abhinavagupta worked from, the one that gave the sequence its name. On the earliest witness it is silent or contrary. That is a real result and it is a narrower one than the summaries of it suggest.

Then the last step. Why those letters on those limbs?

Not because of how they sound. Because of how they look.

Shift attention from the audible phoneme to the written character, he proposes, and the correspondence turns out to be the simplest one available: physical resemblance. The characters that resemble the body parts are those of a Gupta hand of the third and fourth centuries, as classified by Lore Sander from the Berlin Turfan manuscript collection.

NA is the topknot. The four vocalic liquids are the chaplet of skulls worn on her head. THA is the head, CA the third eye and DHA the pair of eyes, the long Ī the nose. BA the mouth, and then the five gutturals are the teeth, running from the right upper fang around to the lower left. The short I the tongue, A the voice. VA the throat. JA the trident in her right hand and RA the trident’s staff, and ṬA the skull-bowl in her left. PA the heart. CHA and LA the breasts, and the long Ā the milk in them. SA the soul, HA the vital energy, ṢA the stomach, KṢA the navel. MA the buttocks, ŚA the genitals, TA the pair of thighs. E and AI the knees, O and AU the shanks. DA the right foot.

PHA is the left foot.

That is abridged. The full list runs to all fifty, and the ears, the earrings, the shoulders, the arms, the hands and the fingers have been left out of the recital above rather than out of the goddess.

His summary of it is worth having whole: the written shape of each character taken singly was identified as resembling a certain bodily limb or organ, and taken together the whole syllabary represented the anthropomorphic body of the alphabet deity. Listed in the conventional head-to-toes order of the installation rite, a particular rearrangement of the alphabet is arrived at, and this is the sequence beginning with NA and ending with PHA. Which means, he adds, that one can draw a sketch of the Goddess using Gupta-era characters.

He then draws it, and captions the drawing as a tentative and provisional attempt.

What follows from that, on his account, is not a mnemonic. He calls the result a synaesthetic icon, and argues that Malini comes into existence through a fusion of phonematic sound and graphemic shape within, or even as, the divinised body of the practitioner. He finds the tradition’s own word for that fusion, lolibhava, a state of indistinctness that things assume when they are on the verge of merging past telling apart. And he suggests what it would have been like near the time the associations were made, when the resemblance was still obvious: the similarity would have been perceived immediately, giving the impression that it was natural, almost inviting the ritual assimilation with the practitioner’s own limbs. Successful realisation of the goddess, he writes, “would induce a mild form of synaesthesia in the initiate”, dissolving the boundaries of sensory perception so that one could see the alphabet as a goddess.

Ben Williams, writing the survey chapter on this whole subject for the Oxford Handbook of Tantric Studies, calls it a breakthrough and an astute decoding. The sequence, he writes, is “not jumbled but rather vertically arrayed to reveal the goddess herself, standing forth with a body of phonemes, offering a visual map for the ritual divinization of the body.”

A reader is entitled at this point to ask what any of these characters actually looked like, because until that question is answered the whole argument is being taken on trust.

So we went and looked.

We opened the tables

Vasudeva’s letterforms come from one book: Lore Sander’s palaeography of the Sanskrit manuscripts in the Berlin Turfan collection, published in 1968 as a supplement to the German union catalogue of oriental manuscripts. He cites it twice and uses nothing else. So far as we have been able to establish, in the nineteen years since his article appeared, nobody writing about his argument has gone to it.

It is not hard to get. The whole volume has been digitised and put into the open repository of the Academy of Sciences at Gottingen with a permanent identifier, two hundred and eighty-two pages, the forty alphabet plates among them, free to anyone. We read it.

Three things came out of that, and the first is small and pedantic and the kind of thing that only matters because of what rests on it.

Vasudeva’s citation of his own source is off by one. He gives the plates as tables 8 to 20 and labels them Schrifttypus II, Sander’s second script type, the Gupta alphabets. Sander’s own list of plates, printed at her page 201, divides them differently: tables 1 to 8 are the Kushana alphabets, her script type I, of about the second and third centuries; the Gupta alphabets of script type II begin at table 9. Table 8 belongs to the earlier type. Nothing in his argument turns on it, and it is worth knowing anyway.

The second thing is what the letters actually look like. We examined them on the plates. Sander arranges each Gupta letter across six hands, three in her Gruppe A of roughly the third and fourth centuries and three in her Gruppe B of the fourth and fifth, so there is no single Gupta form of anything, and which hand you pick changes what you see.

NA, the topknot, is a small compact character, wider than tall. In the Gruppe A hands it is a short stroke across the top with a dip in its middle and, below it and to the left, a thickened tangle, the whole thing trailing off to the right. Sander describes the two parts in her Kushana section, quoting Lüders for the first: instead of the upper horizontal, a weak curve with a notch in the middle; and the lower part of the character written as a closed loop. She says the Gupta form does not differ from the Kushana one, which is why that description carries. On the plate the loop is not open in these hands. She notes that herself, calling it flat and filled in. In the Gruppe B hands the top thickens into a heavier bar and the body below shrinks towards a short stem on small feet, most clearly in the first of the three. A knot sitting on a neck. As a topknot it is plausible, and it is easier to see in the later hands than the earlier.

PHA, the left foot, is the strongest of the three. In the first of the Gruppe B hands it is an upright shank on the left under a heavy wedge, descending and turning right along the baseline into a flat horizontal, with a small closed loop sitting on that horizontal towards the right. An ankle, a sole, and an instep above the sole. It reads as a foot in profile more readily than either of the other two characters reads as what it is supposed to be, though the instep is doing interpretive work and we should say so. In the other two Gruppe B hands the flat baseline is barely there and the loop is a small circle. In the Gruppe A hands it is less settled again, and in one of the three it is not attested at all.

KA, the fang, is the weakest of the three. In the Gruppe A hands it is a vertical stem crossed near the top by a broad bar whose ends droop outward and down, with the stem running well below the bar. Vasudeva’s remark about the earlier form checks out exactly: on Sander’s first plate, a Kushana hand of the second or third century, the character is very nearly a plain plus sign. By Gruppe A it has drawn out into a long blade under a crossguard, which is a fang if you are looking for one, though none of the three Gruppe A hands actually tapers to a point. Two end blunt and one thickens into a blob. By Gruppe B the stem is shorter and tends to hook or loop at the foot, so what little a fang reading had to work with is weaker still. The reading does better in the earlier hands than the later, and it does not do well in either.

There is a problem in that pattern worth naming, and it is ours rather than his. The two readings that fare best here, the foot and the topknot, are clearest in Sander’s Gruppe B, which she dates to the fourth and fifth centuries, while this article has described Vasudeva’s letterforms, after him, as a Gupta hand of the third and fourth. His own citation covers both groups, so he is not caught by this. The tidy formula “a Gupta hand of the third and fourth centuries” is, on the evidence of the plates, too narrow for the argument it is being used to carry.

The fang, meanwhile, is in Vasudeva’s favour in a way that looks at first like a weakness. He already notes that another text read this same letter as a lotus where the compiler of the sequence saw what Vasudeva calls a hideous fang. Standing in front of the plate, both are available, and so is a dagger, and so is a palm tree on a stalk. A character loose enough to be seen four ways is a character a cult could make a fang of.

There is a small gift in Sander’s prose as well, which has nothing to do with Vasudeva and everything to do with the habit of mind his argument assumes. Describing how one scribe formed his KA, she says he wrote the head with its vertical first and then set beneath it the curved crossbar in the shape of a na. A palaeographer in 1968, with no interest at all in Shaiva goddesses, reaching for one letter to describe the drawing of another.

The third thing is about what these plates are, and it is more interesting than the objection we went in expecting. The obvious complaint to make about using a Turfan catalogue to explain a Kashmiri scripture is that the manuscripts are in the wrong place. That complaint does not survive the book. Sander’s chapter containing these alphabets is headed Nordindische Gupta-Alphabete, North Indian Gupta alphabets, and her plates divide at table 28 into Indian alphabets and Turkistani ones, with all the hands used here on the Indian side. She compares them throughout with inscriptions from the Mathura region and, for one hand, with the eastern material at Kaushambi and the Allahabad pillar; of one manuscript she concludes that it was written in Turkistan by a scribe at home in north-west India, or else copied by a Turkistani hand from an Indian exemplar. The manuscripts were found in Chinese Central Asia. The handwriting, on her classification, is Indian, which is the whole reason it can stand in for Gupta letterforms at all.

The real limits are two, and they are quieter. These are Buddhist manuscript hands, overwhelmingly, and the Trika is a Shaiva tradition; nothing establishes that a Kashmiri Shaiva redactor’s letters looked like a Buddhist copyist’s, only that both are in the same broad script family.

And her alphabets are not all of a piece. She explains in a footnote that each was assembled from one or several manuscripts, the fullest supplying the base and the less complete ones filling gaps, and her list at the back shows what that means in practice: of the six hands read above, two rest on a single manuscript each and four draw on two or three. So a column on her plate is sometimes one scribe and sometimes a reconciliation of several, and in a few cells she prints two variant forms side by side rather than choose. Hand-drawn facsimiles rather than idealised type, which she defends as a method against palaeographers who tidy. It does mean that any one shape is a particular hand on a particular leaf, or a small committee of them.

We would add one observation of our own, which is nobody’s argument. The same volume carries later plates that nobody has brought to this question: a late Gupta alphabet of about the sixth century, a Gilgit and Bamiyan type running from the sixth century onward, and a Sharada alphabet of about the thirteenth. Sharada is the script of Kashmir, and the Gilgit and Bamiyan traditions sit far closer to it, in place and in time, than the North Indian Gupta hands of the third and fourth centuries do. Vasudeva’s own position is that the resemblance begins in the Gupta forms and lasts several centuries, so he is not committed to the third and fourth centuries being the only window. Whether the correspondences hold, improve or collapse in those later and nearer hands, we do not know, and we have not done it. Those plates sit a few leaves beyond the ones we read, in a book that is free.

And a last piece of arithmetic that belongs in this section rather than at the end of the article. We looked at three characters. The sequence has fifty. What is above is not a test of Vasudeva’s argument; it is three soundings taken where, so far as we could find, nobody had put a line down, and the other forty-seven are untouched.

What the drawing costs

Vasudeva is harder on his own argument than anyone else has been, and the article that reports him owes the reader his caveats. There are five, and he states them all himself.

It cannot date anything. This is the one he puts first and most bluntly: it would be rash to suppose this could be used as a reliable indication for the date of the earliest Trika texts, because the sequence need not have originated in the Trika, and the range of dates within which the resemblance can be discerned is rather large.

Reading history out of letter-shapes is a poor method in general, and he demonstrates it against himself. He notes that B. Bhattacharya used a comparable argument to date the Jayakhyasamhita as early as 450, and flags that Bhattacharya admitted the inaccuracy of the method. Then he shows why. The Jayakhyasamhita describes much the same characters and reads them in the opposite direction: its author saw the character KA as a lotus, probably with an earlier plus-shaped form in mind, where the author of the scrambled sequence saw a hideous fang, having in mind the dagger-shaped later form of the same letter. The same letters support contrary pictures.

The letters have to be rotated. The Siddhayogeshvarimata says the long Ī pointing downwards is her nose, which Vasudeva observes is meaningless said of a sound or of a nose, and concludes that some of the characters must be turned to fit the icon. His own reconstruction, he says, tries to minimise such rotations. The caption under his figure is itself a hedge: a tentative, provisional attempt.

The textual situation beneath the eleven witnesses is messier than eleven witnesses suggest. The extent of differentiation probably requires many more degrees of separation from the posited original than the known texts allow, so that even the greater part of the known early Trika tantras appear, on examination, to be incomplete remnants of a far more complex textual situation.

And the fifth caveat is the one that most limits what the whole argument licenses, which is presumably why it tends to go missing. He is reconstructing an original meaning and not the exegetes’ meaning, and he says so in a footnote: the post-scriptural commentators built elaborate systems of homologies of their own on this sequence, between phonemes and levels of reality and metaphysical categories, “but here we are concerned with the original attribution of meaning to the phonemes.”

Which means the drawing is not an account of what Abhinavagupta or Jayaratha or Kshemaraja thought they were doing. It is an account of what somebody, several centuries before them, may have been doing when the sequence was first fixed. Those are different claims, and this article has to be careful not to let the first borrow the authority of the second.

Two of the fits are bad, and he says which, though these are notes inside his table rather than items on his list. Of ŚA on the genitals: an apparent problem, since the triangular character for E would clearly be a more appropriate representation, and the resemblance is rather to a male practitioner’s anatomy. Of TA on the thighs: it may originally have been meant to cover both thighs together, since it evolved out of a basic inverted y-shape.

Then there is something that is not his caveat at all. It is the state of the field.

We could not find anyone who has argued with him, and the way we looked matters, so here it is. Book chapters in Indology are badly indexed. The chapter does not appear in Semantic Scholar at all, and the one citation database that might have given a full list was unavailable. What we have instead is a hand-assembled list of four readers: Ben Williams, who calls it a breakthrough; Anna Kavaleuskaya, who uses the body-part mapping in her own work and concludes from the sources herself that the correspondence of letters to bodily parts is “rather standard in many texts,” which is a shade cooler than Vasudeva’s own “no variations at all”; Ana Laura Funes Maderey, who cites it for the phoneme-body explanation, and whose full article we did not read, so we will not characterise her position further; and one doctoral thesis on an aggregator site. Four is not a literature, and absence from an index is not absence of engagement.

Within that limit, two things do seem to be the case. Nobody appears to have redone the comparison against a palaeographic reference work other than Sander’s, or against a different script or a later century’s hand. And until the soundings reported above, nobody else appears to have challenged the step most open to challenge, which is the resemblance itself, judged by eye, by one person, with two of the judgements flagged as weak by the person making them.

The nearest thing to reserve in print comes from an interesting quarter. The volume’s own editors summarised the argument in their preface in the French conditional, and closed by calling the correspondence chose assurément curieuse, assuredly a curious thing. One should not lean on that: French academic prose uses the conditional for reported claims as a matter of course, so it may register nothing at all. The preface is unsigned, and the volume’s editors were Dominic Goodall and Andre Padoux, the author of Vac. If the field’s standing authority on phonematic emanation had a view about whether the Sanskrit alphabet is a drawing of a woman, an unsigned preface is not where he would have put it, and three words in one are the closest we got.

Thirty years of agreement is not thirty years of testing. The invariance the argument rests on is real and checkable in the texts, and anyone can read it. The resemblances are checkable too, as the section above this one found, and until now it appears nobody had bothered.

The nearest thing to a parallel is Buddhist

An article about a differently-ordered alphabet invites comparison with every other tradition that has made letters into a cosmos, and most of those comparisons are ours to make rather than anyone’s to cite.

One is not. The arapacana syllabary is a fixed, non-standard ordering of an Indic syllabary put to religious use, in the same cultural world, some centuries earlier, and in a series of papers from 1990 Richard Salomon argued that its order derives from Gandhari and the Kharoshthi script. That is a real comparandum, established by a careful philologist, and it is structurally the same move: a differently-ordered Indic syllabary whose order turns out to have a historical explanation rather than none. Salomon’s argument is palaeographic and linguistic and the explanation is not the same as Vasudeva’s, so the parallel is in the shape of the problem and not in its solution. We should also say that we identified these papers from Salomon’s own publication list and did not read them. He was not comparing the syllabary to the Shaiva material. So far as we found, nobody has.

The resemblance between all this and the letters of the Sefer Yetsirah, the early Jewish cosmological text in which the Hebrew alphabet is the material of creation, or the Arabic science of letters, is real and it is striking. We could not find a single peer-reviewed study arguing it. Gershom Scholem’s work on the linguistic theory of the Kabbalah is the obvious place to stand if anyone wants to try. Where this essay notices such a thing it is noticing, not reporting.

The man who could have looked, and refused

There is an irony in the material, and the careful version of it is better than the sweeping one.

Abhinavagupta records that an earlier Kashmirian author, Somananda, had treated the shapes of written letters as soteriologically effective in a commentary on the Paratrishika. Abhinavagupta declines to follow him. His stated reason is that such speculations, depending on the artificial conventions of writing, are of no use to disciples removed in space and time, however useful they might be to some rare person.

What that means precisely is that the method was available and was refused. Not that the answer was. Nothing suggests Somananda had the nadiphanta sequence worked out as an anatomy; what the record gives is an interest in letterforms as efficacious, which is the genre of explanation Vasudeva would eventually use and not the explanation itself.

Abhinavagupta’s reason was also correct. Scripts are local and they change, and Vasudeva’s own account of why the principle was forgotten says the same thing from the other side: palaeographic drift and the deliberate invention of new calligraphic styles meant the correspondence was no longer obvious, and the inscriptional record shows it becoming hard to perceive within a few centuries. By the year 1000 the resemblance that had generated the sequence was already most of the way to invisible. A teacher who told his disciples to meditate on letter-shapes would have been handing them the shapes of their own century, which are not the ones the sequence was built from.

So he was right, and the cost of being right was the explanation. The man who declared the sequence principal is the man who ruled out the one line of enquiry that could have said why it was in that sequence, on grounds that were true and that are the very reason it had gone dark.

Some Shaiva tantras did keep the interest. Vasudeva notes the Sarvavira among them, on Kshemaraja’s testimony. The mainstream of the exegesis did not, and the exegesis is what survived.

What the scholars cannot agree about

Three live disagreements, which an honest account names and does not settle.

Are the vowels male or female? This is a flat contradiction in the current literature. Goodall, on the Nishvasa corpus, has the matrika made of feminine vowels and masculine consonants, applied to the left and right halves of the body respectively, which is what makes the Ardhanarishvara costume appropriate. Torzsok, on the Svacchanda, has the opposite: the original layer treats the vowels as male and the consonants as female, with Matrika-Bhairava worshipped by the vowel group and Bhairavi by the consonants, and she says the idea that vowels are masculine and consonants feminine is present throughout the corpus. These cannot both describe one convention. Either the two scriptures genuinely differ, or one reading is inverted. It matters, because Goodall’s whole Ardhanarishvara explanation depends on his polarity and Abhinavagupta’s seed-and-womb scheme depends on the other. For what it is worth, Abhinavagupta’s usage agrees with Torzsok: vowels are seeds, consonants are wombs. That is an observation, not a verdict.

Do the Svacchanda’s alphabet-Mothers belong to the text? Be precise about what is in question. That the text sorts the alphabet into eight groups is secure, and another chapter of it says so. What Torzsok argues was inserted later, following her own doctoral work and Shaman Hatley citing Sanderson, is the passage that gives the Mothers their names, in what became the standard Kashmirian recension. The internal evidence is good: the added lines name seven Mothers where their immediate context has eight, and they assign the first Mother to the vowels, which runs against the gendering of the layer they were put into. Kshemaraja defends the text instead: at the level of Bhairava there are only seven Mothers, so the vowels’ Mother drops out of the count. An unusually clean two-sided case. Philology sees a seam; the tradition’s best reader saw the same crack and supplied a doctrinal reason for it.

Did the Trika’s alphabet theology come from the dualist Siddhanta it claims to transcend? This is the better version of a question usually asked badly. Torzsok proposes that the Trika took Shabdarashi, the male alphabet-god, directly or indirectly from the Saiddhantika Svayambhuvasutrasamgraha or a relative. Her grounds are concrete: he is badly integrated into the Siddhayogeshvarimata; the Svayambhuva names him as the source of verbal creation and compares his power to the wish-fulfilling gem, a simile the Trika texts then reuse; and the affinity between the Svayambhuva and the Trika had already been noticed. Nobody, as far as we could find, holds the view that malini herself was borrowed from outside Shaivism, so that is not a controversy and we are not going to stage one.

Whether the Trika read its own scripture correctly

There is a larger question standing behind all of this, and it needs stating carefully, because it is easy to overstate.

Abhinavagupta opens the Tantraloka with a maximal claim. There is nothing here, he says, that is not indicated in the revered Malinivijayottara by the God of gods, either in its own words or by implication. Of the Lord’s teaching, divided into ten and eighteen and sixty-four scriptures, the essence is the Trika teaching, and the essence of that is the Malini doctrine.

That is precisely the kind of claim philology exists to test, and Sanderson tested it in 1992, in a study of the Malinivijayottara’s doctrine. His finding was that the text gives no certain evidence of the non-dualism Abhinavagupta’s school reads into it, and clear indications of the dualism of mainstream Siddhanta Shaivism, so that, in his words, Abhinavagupta has superimposed the non-dualism of his school on to the Malinivijayottara.

Two things have to be said at once about using that here.

The first is that Sanderson’s subject is not the alphabet. Across some thirteen thousand words his article does not mention matrika, malini as an ordering, Shabdarashi, or the split and undivided wombs even once. It cannot be cited on the two orderings. If a source cites it for them, the source is wrong.

The second is that what it does establish is a habit. Sanderson’s charge is that the tidy doctrinal system Abhinavagupta attributes to his root scripture is partly his own construction, imposed on a text that was doing something else. And the tidying of the two alphabets, from the scripture’s untidy pair of Malinis into a clean opposition with a derivation connecting them, looks like the same hand at work. That connection is this essay’s and not Sanderson’s. He argues the pattern on other evidence; we are suggesting this is another case of it.

It should also be said, since this is a place where accounts flatten things, that Sanderson concedes what can be conceded. He identifies the two places in the whole Malinivijayottara where non-duality is mentioned at all, says it is very doubtful that monistic non-dualism is intended in either, and then adds that this sort of non-duality is not inconsistent with the ontological non-dualism Abhinavagupta sees in the text. He is reported to have allowed later that the eighteenth chapter could be read as enjoining non-dualistic awareness, though that comes to us through a doctoral thesis quoting a different publication of his, and we have not read it. The man who reported the superimposition was careful about the gap he was reporting.

And the question is not his alone, which an account naming only him misrepresents. Jurgen Hanneder’s study of Abhinavagupta’s own verse commentary on the Malinivijayottara’s opening is the principal work on the hermeneutic by which he gets his Trika out of the text. Vasudeva’s edition of the yoga chapters is where the scripture’s system is set out on its own terms. We read neither. Anyone who wants the argument rather than its headline should go to those two, and to Sanderson’s own pages, rather than to this paragraph.

Who still puts letters on a body

The modern question is where this operation is alive, and the honest answer is more interesting than either of the easy ones.

It does not appear to be part of ordinary brahminical domestic worship, though the evidence for that is narrower than the claim usually made from it. Gudrun Buhnemann’s study of smarta ritual, the mainstream brahminical observance that takes the Veda rather than the tantras as its authority, does have an installation rite in the daily sequence. It is not the alphabet. It is the sixteen verses of the Purusha Sukta, the Rigvedic hymn of the cosmic person, placed on sixteen limbs, with shorter five- and six-limb forms more common in current practice. The scope to keep in mind is that her base text is one Maharashtrian manual of the Rigvedin tradition, read alongside the legal digests, so this is a finding about that tradition and not a census of India.

Her own notes on the rite are more useful than its contents. The tradition of installation, she notes, “seems to be of Tantric origin”, and “as part of the smarta puja nyasa is not performed by all.” Then she quotes an appendix to one of the domestic manuals rejecting the rite outright. Some do not want it, for that rule is not Vedic. In a later book she puts the whole situation in one clause: the installation rite forms part of current smarta worship and is rejected by some as Tantric.

A rite inside the tradition that part of the tradition considers foreign to it.

The alphabet does enter her material once, in the rite for infusing life into an image. The version she edits is specifically the one performed without installation on the bodies of the worshipper or the image.

Nor is it in the daily liturgy of the Shaiva Siddhanta, the dualist Shaiva school with its own canon of scriptures, which holds that souls remain distinct from Shiva for good. Its standard South Indian ritual manual is the Somashambhupaddhati, and we could not get Helene Brunner-Lachaux’s edition and translation of it, which is still in print and not legitimately available in any digital form. Going instead to the Kriyakandakramavali, a Sanskrit text of the same tradition printed at Devakottai in 1934 with a Tamil translation, one finds a great deal of installation: the five Brahma-mantras, the six ancillaries, the levels of reality, the kalas or graded portions of the godhead, the Shiva-hand, the knowledge-body, the image, the weapon, the linga, the pedestal, even the foundation stones of a temple. Of the alphabet goddess, nothing. The fifty letters are present in that rite, and they are the filaments of the lotus throne that Shiva sits on, not marks on the worshipper. That finding rests on a search of a Tamil translation’s optical character recognition and should be treated as strong rather than settled; the thing that would make it certain is Brunner’s volume on a library shelf.

And then the fact that changes the scale of all this.

Matrika-nyasa is a named, obligatory step of the Bengali Durga Puja.

Hillary Rodrigues’s liturgical ethnography of that festival as performed in Banaras documents it three times over: an internal installation of the Sanskrit alphabet upon the officiant’s transformed body, an external placement of the syllables after it, and then a samhara installation, the same procedure in reverse, by which he enfolds the goddess’s vibrational body into his own. The officiant takes refuge in the deity of speech whose body is composed of the fifty Sanskrit syllables. The rite recurs in the worship of the young girl as goddess. It is in his glossary as a standard term of the liturgy.

And Rodrigues independently confirms the point about smarta ritual, citing the same page of Buhnemann: installation does not always form part of smarta or Vedic worship. He calls the rite likely of Tantric origin, and names the Durga Puja’s Tantric character as the prime feature distinguishing it from the non-Tantric brahmins’ worship of the Goddess.

Which gives the honest formula. This is a tantric rite. It is absent from the Vedic and smarta core, and absent from the Saiddhantika daily office. And it is performed in public, by ordinary hereditary priests, on behalf of patrons who have been initiated into nothing at all, inside the largest annual festival liturgy of Bengal. Tantric does not have to mean esoteric.

Rodrigues watched it in Banaras, among the Bengali community there, at a public marquee and in a family’s front room.

Elsewhere the record thins and we will say so.

In Nepal there is a man with a name. Jeffrey Lidke watched Bajracarya Sharma, priest of the living girl-goddess of Patan, at the daily worship he had performed for more than twenty years: seated in posture, breath regulated, building the diagram inwardly and meditating on its centre, and then installing the beings and powers of that internalised diagram in his own limbs. Encoding his body, in Lidke’s word. Only after that did he open his eyes and receive the goddess’s gaze. The phonemic theology is explicit in his lineage, which understands the Goddess as manifesting through her phonemic self. But what Lidke watched was the installation of a diagram and not of the alphabet, and we are not going to claim more than he saw. The cult of the goddess Kubjika, which belongs to the Western Transmission and is still practised in the Kathmandu valley, is the obvious place to look next, and we have not looked. Note that its scriptures are already among Vasudeva’s eleven witnesses, which transmit the NA-to-PHA sequence itself; whether that lineage also uses an alphabet code of its own is a question we are raising and not answering.

In Srividya it is attested, in Douglas Brooks’s study of the South Indian traditions. It should not be assumed universal there even so: lineages differ about what may be placed on a body and where, and we have not surveyed them. On Kashmiri practice today we have essentially nothing, and that is a hole in our record rather than an absence in the world.

Pancharatra Vaishnava texts appear to carry alphabet-installation chapters of their own, one of them titled for the rite. We have that from a catalogue description rather than from the texts, so treat it as a lead.

The alphabet is the name of a script

One last thing, which nobody set out to arrange.

A great many Sanskrit manuscripts and inscriptions of roughly the sixth to twelfth centuries are written in a script that works on Indian epigraphy sometimes call Siddhamatrika. It is a later hand than the one Vasudeva reads, and the two should not be run together; what they share is the name. The name means the Siddha alphabet: matrika, as ever, is simply the word for alphabet. The earliest record of the Indian script being called that comes from al-Biruni, around the tenth century, which is to say in Abhinavagupta’s own lifetime.

Its East Asian descendant is called Siddham, Shittan in Japanese, and it is the script in which Sanskrit seed-syllables are written and venerated in Shingon Buddhism. Anyone who has looked at a bonji, one of those single Sanskrit seed-characters painted on a Japanese temple plaque or carved into a grave marker, has met something called the Matrika without being told.

One small convergence, which we flag as ours and nobody’s argument. The script’s other name, from its letter-forms, is kutila, bent or curved. And kutila is what Torzsok finds the alphabet goddess called in her coiled aspect in the Nishvasa Guhyasutra, the upward-going Kutila, the most subtle form of sound. The same word names the coil of the goddess and the curve of the letters. No scholar has connected them and we are not pretending one has.

The invocation om namah siddham survived at the head of alphabet primers into the nineteenth century. Which is the furthest downstream this material goes. A child’s first page.

Beyond that the cultural afterlife is close to empty, and the reason is worth stating rather than papering over. We searched for her in modern art, film and popular culture and found nothing at all. Whether any surviving sculpture or painting depicts an alphabet goddess, as against the Seven Mothers who share part of her name, we could not establish either way, so that is an open question and not a negative finding.

What the texts do say is that she went the other way from everything a cult usually does. In the later Trika, Vasudeva notes, Malini becomes wholly aniconic, transcending visual representation altogether. The one place elaborate standard images of her are found is the Kubjika cult, and he suspects that elaboration followed a demotion: she got a picture when she stopped being the highest thing.

So where does that leave the thing we began with.

A sequence of fifty sounds, taught in scripture, used to letter a body and to draw mantras out of it, declared principal by the most formidable reader the tradition produced, and never explained by him or by anyone after him. One modern argument says the reason is that the characters look like the parts they name. It holds for the version Abhinavagupta worked from and not for the oldest one we have. It rests on resemblances one scholar saw by eye in one reference work, which so far as we can establish nobody had gone back to look at until we did. Of the three characters we checked, the foot is good, the topknot is fair, and the fang is thin in the earlier hands and thinner in the later. Forty-seven remain, along with a shelf of nearer scripts nobody has tried.

What is not in doubt is the thing underneath it, and that is the stranger fact. For several centuries, across texts that disagreed about nearly everything else, under redactors who reshuffled the order freely, almost nobody moved a letter off its limb. Whatever the fifty sounds were, they were somewhere in particular.

The medium was never an image.

It was a body, and bodies are not kept.