The philology of unintelligibility

Añjanīnandan Śaraṇ gave a large part of his life to one poem.

His work was to gather the named commentarial tradition on the Rāmcaritmānas, verse by verse, and set the readings beside one another, so that a reader could watch the whole tradition argue in one place. The Gita Press printed the result at Gorakhpur in 1968, under the title Mānas-pīyūṣa. It is still the book a serious reader of the Mānas reaches for.

Then he came to the half-line this chapter is about, a short concession in the opening invocation of the Bālkāṇḍ in which Tulsīdās grants a charge against the śābara mantras rather than fighting it. Śaraṇ stopped. He laid out four readings of it.

He chose none of them.

That is the fact to open on, and the reason it gets missed is that one of the four readings is very good indeed.

That reading is this. Anmil ākhar, ill-matched letters, is a judgement about register rather than about meaning. The syllables are badly assorted. They do not sound the way sacred speech is supposed to sound, and the complaint is about how they are dressed rather than about what they carry. It is a good reading. It is also one of four readings a Hindī commentator had already set out on a single page, three of them attributed by siglum rather than by name, in print and on the shelf since 1968.

And one of the rivals on that page says the opposite. It reads the line as meaning that the letters and the sense do not match, and then draws the inference out loud: these mantras are not devoid of meaning. The meaning does not tally with the letters.

So the question this chapter exists to settle was never settled by the tradition either. Four readings, one page, one commentator, no verdict.

A treatise that wanted to announce a finding has to report a disagreement instead, and take a side while naming what it is taking a side against. The reading this chapter follows is the register one. It follows it on stated grounds and not on authority, and it does not pretend the rivals are weaker than they are.

There is a second thing standing between a reader and this half-line, and it comes from the opposite direction. When a reader in English meets syllables that will not parse, a famous theory is already waiting for them, and it was written by a man who had gone to watch a fire altar being built.

The claim everyone reaches for

In 1975, in a village in south-west India, Nambudiri brahmins performed the Agnicayana, and a Dutch philosopher of language was there to record it. Frits Staal had spent a long stretch of his working life on the Vedic recitational corpus. He came away from it with an argument.

The argument is that ritual has no meaning.

Not that its meaning is obscure. Not that its meaning has been forgotten. That there is no meaning there to find, in principle, because ritual is rule-governed activity of a kind that does not refer to anything outside itself. His 1979 article in Numen builds towards the hypothesis that ritual has no meaning, goal or aim, and then states it flat: “Ritual is pure activity, without meaning or goal.” Later in the same article he turns to language: “Ritual is replete with language, but it is very often meaningless language.” He then prints four lines of syllables chanted over a golden image buried under the fire altar of the Agnicayana, and it is of those printed syllables, and not of meaningless language in general, that he writes: “Such structured sounds partake of the syntax of ritual, but do not relate to meaning.” The next sentence widens it himself: this applies to most mantras. Syntax without semantics. Structure that behaves like grammar and carries no message. The position travels under the phrase Staal gave it in that article’s title: the meaninglessness of ritual.

It is a serious argument, made by a serious scholar, and it is the wrong tool for this half-line.

One declaration at the outset, because it governs how everything below is written. One text of Staal’s has been read for this treatise and it is the only one from which anything here is asserted: “The Meaninglessness of Ritual,” his 1979 article in Numen, quoted above and the source of the Agnicayana date and place. His monograph and his contribution to the Alper volume have not been opened, and nothing is claimed from either. Where other unopened books are named below, no proposition is attributed to any of them.

Go looking for anything written in English about mantras that do not parse, and Staal arrives quickly. He is the most cited English-language treatment of ritual meaninglessness, and in the popular and semi-scholarly material surveyed for this treatise he is routinely applied to anmil ākhar.

That is a claim about what this research read, and it is as wide as this research can make it. A historiographical absolute about the whole English-language conversation since 1979 would need a survey of the literature, and one article of Staal’s is what this chapter has read. No such absolute is asserted anywhere here.

There are good reasons Staal is reached for. He was working on the hardest available case, the Vedic recitational corpus with its inserted syllables that sit inside the chant doing structural work and refusing to translate. He knew Sanskrit. He knew formal linguistics. His answer was not a shrug. It was a thesis: that the search for meaning is a category error imported by observers who assume anything language-shaped must be saying something.

So when a modern reader meets a body of vernacular syllables that resist parsing, Staal is sitting right there with a finished explanation.

What the commentators did with the half-line

Return to the verse from chapter 1, the four lines out of the Bālkāṇḍ invocation that the popular literature quotes as its warrant. The half-line at issue is the concession:

अनमिल आखर अरथ न जापू।

anamila ākhara aratha na jāpū Ill-matched letters; no settled meaning, and no japa.

Rāmcaritmānas, Bālkāṇḍ 1.15.3 · first half of the second line only, not a complete chaupāī #

Anmil ākhar. Ill-matched syllables. Anmil is the negative of mil, to meet, to join, to match. It is the word for cloth that does not match, for a pair of things that do not belong beside each other. On its own it is a word about assortment.

Stopping there, and declaring on the strength of the word alone that Tulsīdās was passing a judgement about register and nothing else, will not do. There are two reasons, and this treatise has both of them in its own files.

The first is internal. Chapter 1 prints the Gita Press gloss on the line, which says the letters are mismatched, that there is no proper meaning, and that there is no japa either. Chapter 8 reports the same gloss and calls it blunt about there being no proper meaning. A chapter cannot call the reading of the line as a claim about meaning the very error it exists to correct while two other chapters of the same treatise report its own cited authority making exactly that claim.

The second is external, and it is the more interesting of the two. Śaraṇ’s Mānas-pīyūṣa is the standard Hindī compendium of Mānas commentary, collating the named ṭīkā tradition verse by verse. On the page it gives this half-line it sets out the construction of the words, in its own phrase, in several ways, and it gives four: the base gloss chapter 1 already has, with an alternative offered inside the same sentence, and three lettered construals after it, each attributed on the page by a siglum rather than by a name.

They do not agree with each other. One of the three says outright that the śābara mantras are not devoid of meaning, and that the meaning simply does not tally with the letters. Another reads the letters as unjoined and lists, in technical terms, the rules of composition they breach. The unjoined-letters reading is a verdict on how the syllables are dressed. The other is a claim about what they carry, and it is a plain claim about meaning.

A Hindī commentator, in a Gita Press compendium, sitting down and listing the specific rules the syllables break. That is the reading this chapter takes, and it takes it knowing the same page holds a commentator who went the other way.

The sigla by which Mānas-pīyūṣa attributes all three lettered construals are expanded in the compendium’s own table of sigla, and that table’s columns are scrambled in the scan worked from here. All three are therefore attributed to Mānas-pīyūṣa reporting unexpanded sigla, and none of them to a named individual, until a printed page can be read. Chapter 8 reports the same page and enumerates and attributes it the same way.

So the register reading is real and it is Indian and it is old. It is also one of four, counting the base gloss with its embedded alternative as one and the three lettered construals as three, and at least one of the four is a plain semantic claim. The distinction is in the commentarial tradition already, and the certainty a modern reader might want to attach to it is the one thing the tradition does not supply. What this chapter has is a reading with its rivals named.

Why a decorum claim is what you would expect here

Because of who is speaking, and chapter 1 established who.

The verse is not in a ritual manual. It is in the maṅgalācaraṇ, the opening invocation, in the poet’s own voice, in the passage where he is arguing for his right to write the life of Rāma in Avadhi. He is about to make a book out of vernacular speech in a literary culture that reserved authority for Sanskrit. He needs a precedent for potent language that is linguistically irregular.

So he reaches for a category of mantra everybody already granted was effective and everybody already knew was linguistically irregular, and he grants the irregularity in full. The syllables are ill-matched. Yes. They are. And they work anyway, by Maheśa’s power.

A poet building that argument has every reason to make a claim about composition and no obvious reason to make a claim about semantic vacuity. His own poem means something. A claim that these syllables mean nothing in principle would argue him out of his own case.

That is the strongest argument for the unjoined-letters reading, and it is an argument from the passage rather than from a theory. It is also not decisive, because the rival construal has the commentator’s own inference behind it and reaches nearly the same place by a different road: the meaning is there and does not sit on the letters.

What both of them are not is Staal. Staal’s claim is metaphysical. It is a claim about what ritual language is, everywhere, as such, whether or not anyone objects to the sound of it. Meaninglessness in Staal is a structural property of the thing. Ill-matchedness in the Mānas commentators is a verdict passed by listeners with standards, and they itemise the standards.

One of those is a statement about the nature of language. The other is a review.

The register was chosen

There is a further move available here that this treatise had in its own dossier and never made, and it comes from Indian scholarship.

Hazārīprasād Dvivedī, in the third edition of Hindī Sāhitya kī Bhūmikā, published in 1948, reads the Nāṭyaśāstra’s handling of śabara, ābhīra and cāṇḍāla speech, and Daṇḍin’s remark in the Kāvyādarśa that in poetry the speech of the Ābhīras and the like is called apabhraṃśa, and takes both as evidence about literary language rather than about dialect. Chapters 2 and 7 cite the Nāṭyaśāstra passage at 17.49 and Dvivedī cites it at 17.48; that is a recension difference and the treatise must state which numbering it follows.

Dvivedī’s framing does the work Staal cannot. If the irregular register is a literary register, chosen and marked as such, then unintelligibility is a property of a decision rather than of a defect.

And “marked as such” is now a measurable statement rather than a figure of speech.

Chapter 10 ran the same four-term count over two occult books from the same bazaar shelf that are not Śābara compilations: Bṛhad Indrajāl, Hāthras 1964, and the Khemrāj Uḍḍīśatantra. On the four words that mark the Śābara register, both of those books come in at or near nothing, and the Śābara compilations come in one to two orders of magnitude above them. Same shops, same customers, same subject, and a lexical gulf between them.

A register that separates that cleanly from its nearest neighbour is a register somebody is choosing. Whatever else anmil ākhar is doing, it is describing a body of language that a lexical test can pick out of the shelf it sits on, which is what Dvivedī’s argument would predict and what a theory of ritual language as such would have no reason to expect either way.

That is a lexical separation and it establishes nothing about intelligibility. A register can be distinctive and still mean something, or be distinctive and mean nothing. The controls show that the vocabulary is specific. They do not show that the syllables were chosen for their sound.

And there is an older statement of the same thing from a different tradition entirely. Shashibhusan Dasgupta, in the first edition of Obscure Religious Cults (University of Calcutta, 1946), reports from the Vimalaprabhā, the great commentary on the Laghukālacakratantra, that the doctrines of the pāramitā way are set down wholly in Sanskrit while those of the mantra way are explained in Sanskrit, in Prakrit, in Apabhraṃśa and in non-Sanskritic speech, with the Śabaras named as the example. Dasgupta read that in manuscript, and this treatise has not opened the printed edition that now exists at Sarnath. The passage is held through Dasgupta and reported as his.

If that holds, an eleventh-century Buddhist Tantric commentary states as a principle that mantra-teaching is delivered in non-Sanskritic speech. The syllables are irregular because of where they were deliberately put, and the irregularity is the point.

Unintelligibility as the observer’s verdict

There is one more thing worth saying about this word, and it comes from a book chapter 14 is built on.

Verrier Elwin, in the Preface to The Religion of an Indian Tribe (1955), surveys everyone who wrote about the Saoras before him. Fred Fawcett, the Ganjam police superintendent whose 1888 lecture gave what Elwin calls the first detailed account of the Saoras of Ganjam District, comes out of that survey well: Elwin rates him as an observer who took care and got things right. Then he enters one qualification, in the manner of his time. Where a chanted formula defeated him, Fawcett reached a little too quickly for the word gibberish. Elwin 1955 is in copyright and no passage of it is quoted here. The one word is named, because which word he used is the whole of the point, and the word on the page is gibberish. The passage sits in the Preface’s survey of previous writers, and the source note records what still has to be checked against a printed copy.

That is the whole of this chapter’s problem in one clause, from an ethnographer who saw it happening in his own discipline. Unintelligibility is very often the observer’s verdict rather than the material’s property. Fawcett could not understand the incantations and so classified them as gibberish. A modern reader cannot parse the Śābara syllables and reaches for a theory that says they were never parseable. The move is the same move and it is a century apart.

The apparatus this chapter cannot supply

If Staal is the wrong tool, something should be the right one, and this treatise is not in a position to say what.

Robert Yelle’s Explaining Mantras (Routledge, 2003) is the obvious candidate on its title and subject, and it has not been opened here. Nothing is attributed to him and no ranking is made between his frame and Staal’s, because this treatise cannot rank an argument it has not read, and it does not know what Yelle says. What can be said is what a reader would go to him for: a rhetorical account of mantric language, and a treatment of the Tantric aspiration towards a language in which sound and referent are not arbitrarily linked. Whether his account supports anything in this chapter is unknown.

Alper’s edited volume Understanding Mantras (1989) is the standard collection in which the debate about mantric meaning sits in one place, and it too has not been opened. It would say which arguments have already been run against Staal, and whether the register-and-semantics distinction drawn here has been drawn before. It may well have been. This chapter cannot say, and it therefore does not claim the distinction as original.

The closest available comparison, the Buddhist dhāraṇī material carried out of India into China, Korea and Japan, is the comparison this chapter should be making and cannot. Paul Copp’s The Body Incantatory (2014) is a title in this treatise’s bibliography and has not been opened. Nothing is claimed here about whether the dhāraṇīs were transliterated or translated, because no page-level source for it has been read. The comparison is named as unmade.

A hypothesis of this treatise, marked as such

What follows is a hypothesis of this treatise. It is not a sourced claim. No scholar named in this chapter asserts it, and it should never be cited as though one did.

Unintelligibility is often a transmission property.

Syllables that no longer parse are, very frequently, syllables that crossed a language boundary, or an oral century, or a scribal hand that did not know the language it was copying. They kept their shape and lost their sense, and the loss is an accident of the road rather than a design feature. The Śābara material is by its own charter multi-lingual, proclaimed in five named vernaculars in chapter 3, and the printed compilations chapter 12 counts reach across the Hindu-Muslim line: five of its eight books name the prophets, one of them thirty-three times. One whole series of them, chapter 11 shows, was assembled from material posted in by named practitioners in at least seven states. Material gathered from that many mouths, transmitted orally, then printed cheaply by presses that were not doing philology, is exactly the kind of material in which words go opaque without anyone intending it.

If that is right, anmil ākhar may be describing the visible residue of transmission. Ill-matched, because the pieces came from different places and were joined by the road rather than by an author.

And here is what would test it, stated so that the hypothesis is falsifiable rather than merely attractive: a critical edition with a full apparatus, variant readings collated across witnesses, the points of corruption visible on the page. Opacity that is transmissional would show itself as instability across witnesses. Opacity that is designed would sit stable across all of them.

No such edition exists, and chapter 17 explains why: the material reached print through a commercial bazaar circuit that had no reason to make one, and it has attracted no philologist since.

The witnesses for that test are on the table.

Chapter 10 has established that eight independently compiled printed books exist in machine-readable form, from six houses in three cities, and that they are neither one text reprinted nor eight unrelated compositions. They are a partially shared stock. Three of them hold a substantial quantity of running text in common with the big Chaukhambā volume of 2016, while the Delhi paperback this treatise started from shares almost nothing with any of them and is the odd one out of the eight. The overlaps, book by book, are in that chapter.

Shared material across independent witnesses is precisely what a collation requires. Two books that share nothing cannot be collated. Two books that share everything are one book. A quarter of one book’s running strings appearing in another, printed in a different city by a different house, is the condition under which the question this chapter asks can actually be put to the material: take a passage that stands in three or four of these compilations and look at whether the opaque syllables inside it stand still.

Nobody has looked.

The overlap figures in that chapter are floors and not estimates. Optical noise depresses every one of them, so the real sharing is higher than the numbers say, and how much higher is unknown. They also say nothing about which book borrowed from which, or whether all of them draw on something earlier that nobody has found. They establish only that the material for a collation is there and that it is freely downloadable.

Chapter 11 supplies one live instance of exactly the instability the test would look for: a sequence that stands thirty-three times in the 2016 compilation is answered in the Gita Press commentary’s specimen by ईश्वर उवाच, an ordinary Sanskrit speech-frame, and this treatise cannot say which is older. One instance is an anecdote. Eight witnesses with a quarter of their material in common is the beginning of an apparatus.

So the hypothesis stays a hypothesis. It is a testable one, and it is testable because eight books can be collated where one book cannot.

What this chapter has and has not done

The ledger, plainly. Staal’s 1979 Numen article has been opened and is the one text of his quoted here. Yelle, Alper and Copp have not been opened and nothing is asserted from any of them. The comparative dhāraṇī case is unmade. The frequency material is not used here to stage a contest between two theoretical frames, because that contest has no losing side, and this is the cleanest illustration this treatise has of the point made a few lines below about orders of claim. Eight printed compilations now stand behind those counts, with two negative controls beside them. Eight times the evidence, and not one sentence of the Staal argument moves. Scope is not the reason. He is not confined to the Veda: the sentence immediately after the second one quoted above widens the claim in his own words, to most mantras. What puts him beyond the reach of a word count is the order of the claim. A metaphysical proposition about what ritual language is, as such, does not have a frequency, and nothing counted in a paperback could have gone against it. What is left is a reading of a half-line, four construals of it from the Indian commentarial tradition, and one distinction that holds however the construals fall.

Anmil ākhar is a verdict on how the syllables are put together. It is aesthetic and social, and the commentators itemise the rules being broken. Staal’s claim is that ritual has no semantics. It is metaphysical and universal. Anyone who cites the second to settle the first has substituted a twentieth-century theory for a sixteenth-century judgement, and has done it without noticing, because the two sound alike in English translation and only one of them is famous.

The next chapter leaves the philology and goes to the hills of southern Odisha, to the people whose name this entire category wears.

The syllables were never asked to make sense.

They were asked to be well-dressed, and a Hindī commentator sat down and listed the ways they were not.