Meaning Uncertain: The Rigvedic Words Nobody Can Translate

One verse, three unknowns

Near the end of a Viśvāmitra hymn to Indra, in the third mandala, there is a verse that any Vedic reciter in India can still deliver with every pitch accent in place, and that nobody on earth can fully translate.

Among the Kikatas what do thy cattle? They pour no milky draught, they heat no caldron. Bring thou to us the wealth of Pramaganda; give up to us, O Maghavan, the low-born.

(RV 3.53.14, Griffith 1896) [1]

Griffith’s rendering looks confident. It is not. Three of the words in that line occur exactly once in the entire Rigveda, and twenty-five centuries of readers have failed to agree on what any of them means. Kīkaṭa (कीकट) is a people, or a place, or possibly neither. Pramaganda is a person, or a title, or an insult. Naicāśākha, which Griffith turns into “the low-born”, is a compound whose parts are clear and whose sense is not: something about a “low branch”, attached to somebody, for some reason.

I checked the occurrence counts myself against the Saṃhitā text rather than trusting a summary: each of the three stems appears in one verse of the corpus, and it is the same verse. Lubotsky’s concordance lists every word form of the Rigveda with its frequency and the text of every pada in which it stands, which is where a count like that can be checked in print. [2] One line of poetry carries three dead words.

What makes this worth more than a footnote is the contrast with everything around it. Among long texts surviving from antiquity, the Rigveda has the strongest claim to near-perfect transmission. The recitation schools carried it across roughly three thousand years with its phonemes, its syllable count, and its musical pitch accent intact, through a period when nobody wrote it down at all. That machinery, described in more detail in the piece on the oral engine behind the text, protected the text’s sound with an obsessiveness that has no parallel. [3] It did not protect the text’s sense, because sense was never what the machinery was built to carry.

This piece is about that gap: the vocabulary of the Rigveda that is perfectly preserved and genuinely unreadable, the two and a half thousand years of attempts to read it anyway, and what the whole history tells us about how meaning actually survives.

A dictionary before there were dictionaries

The oldest surviving response to the problem is also the oldest surviving Indian linguistic text. Somewhere in the middle of the first millennium BCE, a scholar named Yāska wrote the Nirukta, a commentary on an older word list called the Nighaṇṭu. The Nighaṇṭu is not a dictionary of Sanskrit. It is a dictionary of the Rigveda’s hard words, which is a very different object, and the difference tells you something.

Its first three chapters collect roughly 1,341 words in semantic clusters: words for physical things, then words for the body and human qualities, then abstractions. The fourth chapter, the aikapadika section, is the interesting one: it gathers words that are ambiguous or singular, the ones a reciter would stumble over. The fifth lists divine names sorted by region of the cosmos. [4]

Yāska is explicit about why such a list had to exist, and his explanation is the earliest statement of textual loss in the Indian tradition:

Seers had direct intuitive insight into duty. They by oral instruction handed down the hymns to later generations. Later generations, declining in the power of oral communication, compiled this work, the Veda and the auxiliary Vedic treatises, in order to comprehend their meaning.

(Nirukta 1.20, trans. Lakshman Sarup, 1921) [5]

Read that twice. The author of India’s first work of philology is telling us, at the start of his book, that the people who made the hymns understood them and the people who inherited them did not, which is why he has a job. Whatever one thinks of the theology wrapped around it, the philological observation is exactly right, and it was made more than two thousand years before the first European read a line of the Rigveda.

Aside. Yāska’s method was etymological: break the hard word into roots, and derive a meaning from what the roots suggest. It produced some brilliant results and a great many bad ones, because the technique has no brake. Any string of Sanskrit syllables can be decomposed into something, and a derivation that sounds plausible cannot be distinguished, by the method itself, from one that is true. The modern descendant of this problem is still with us: a reading that explains the verse is not the same as a reading that is correct.

“The mantras are meaningless”

What makes the Nirukta remarkable is that Yāska does not begin from the assumption that the hymns can be understood. He begins by reporting an opponent who denied it.

In Nirukta 1.15 and 1.16, Yāska sets out the position of one Kautsa, who argued that the Vedic mantras are anarthaka, without meaning or purpose. Kautsa’s case, as Yāska reports it, runs through several arguments: the mantras are recited in a fixed order that no ordinary utterance requires, their ritual function does not depend on anyone understanding them, parts of them appear to say impossible or contradictory things, and, crucially for our purposes, they contain words found nowhere else in the language. Yāska’s reply is that the Veda uses the same words as ordinary speech, and that it has to be studied as a system, with its own internal evidence brought to bear. Paolo Visigalli and Yūto Kawamura, who published the most recent annotated translation and analysis of the passage, treat the exchange as the turning point at which Vedic interpretation became a systematic discipline rather than a matter of ritual competence. [6]

The structure of that ancient argument is worth holding onto, because it recurs everywhere. One side says: the text’s obscurity shows that meaning was never the point. The other says: the text’s obscurity is a problem to be solved with evidence. Neither side wins outright. Kautsa is wrong that the hymns are meaningless, as any reader of the Uṣas hymns can see in ten minutes. He is right that the ritual tradition preserved them without requiring their meaning, and that some of their words had already become noise.

Why the sound outlived the sense

The reason for the asymmetry is mechanical, not mysterious.

The recitation tradition encoded the hymns as sound and checked them as sound. The saṃhitāpāṭha preserves the continuous text with its sandhi; the padapāṭha resolves it into separate words; the permuted recitations (krama, jaṭā, ghana) scramble the word order into interlocking patterns that make a slip audible. Each of those is a form of error correction on the phonetic string. None of them is error correction on meaning. A reciter can carry a word perfectly for fifty generations while the thing it referred to disappears, and nothing in the system will register an error, because no error has occurred.

The padapāṭha deserves a line of its own, because it solves a problem most readers of translations never see. The Rigveda as recited is a continuous stream in which word endings and beginnings fuse. Split the stream wrongly and you get a different sentence.

Methods note. To get a feel for the scale of that problem I ran a count over the Saṃhitā text in this project’s own data files: 130,048 whitespace-separated tokens across 10,528 verses, of which 47,357 forms are distinct and 35,481 occur exactly once. The 74.9 per cent singleton rate is not a measure of rare words. It is mostly a measure of sandhi: the recited text chunks words together, so every new combination counts as a new form. That is precisely the artefact the padapāṭha was built to remove, and the number is a decent way of appreciating why an analytical recitation had to be invented. Alexandre Sotov’s corpus study, working with properly segmented data, puts the Rigveda at about 165,000 word tokens and finds that hapax rates vary systematically with a hymn’s subject matter, which is a more disciplined version of the same observation. [9]

There is one honest piece of arithmetic to extract from a count like mine. In the simplest Good and Turing estimate, [25] the probability that the next token drawn from a population is of a type never yet seen is approximated by the proportion of types seen exactly once:

$$p_{\text{unseen}} \approx \frac{N_1}{N}$$

With $N_1 = 35{,}481$ and $N = 130{,}048$, that is roughly 0.27. Taken at the level of phonetic chunks rather than lexemes, it says something stark: if another Rigvedic hymn surfaced tomorrow, around a quarter of its word-shapes would be new to the corpus we have. We are not reading a language. We are reading a sample of one, and the sample is the only informant left.

Three words, one verse: an anatomy of not-knowing

Return to RV 3.53.14 and watch what happens when the tradition meets a word with no second attestation.

For kīkaṭa, Yāska states that it is the name of a country, a non-Āryan one (Nirukta 6.32). Later Sanskrit treats kīkaṭa as a synonym for Magadha, which is how the word ends up on modern maps in Bihar, several hundred kilometres east of anywhere else the Rigveda describes. The chain is worth separating out: a hapax, an ancient gloss, a later lexical equation, and then a geographical conclusion drawn from the equation rather than from the verse. [10]

For pramaganda, Yāska offers “son of a usurer”. Macdonell and Keith, collecting the scholarship in the Vedic Index, record the gloss and dismiss it as hardly probable. The name has no convincing Indo-Aryan analysis, which is one reason it has been drafted into the long argument about non-Indo-Aryan vocabulary in the Rigveda, a debate with its own literature and its own rather different numbers. [11]

For naicāśākha, the traditional reading takes it as an epithet of Pramaganda, “belonging to a low branch”, that is, of poor family. Alfred Hillebrandt proposed instead that it describes the soma plant, nīcāśākha, “with shoots turned downwards”, making the verse a complaint about people who keep neither the milk cult nor the soma cult. [11] [24] That is not a small adjustment. It changes the verse from a raid on a despised chief into a statement about ritual difference.

Notice what the table actually shows. These are not corrupt readings. The words are metrically sound, accented, and transmitted without variants. The failure is referential: we have the signifier in excellent condition and no access to the world it pointed at. The frog hymn’s akhkhalīkṛtya is the happy case, because frogs still croak and the imitation is self-evident, which is why that word has become a favourite example of Rigvedic humour. The Vṛṣākapi dialogue is the unhappy case: twenty-three verses of domestic comedy whose jokes depend on a situation nobody can reconstruct, which has not stopped it from being one of the most discussed hymns in the corpus.

The hymn where translation visibly broke

If you want to see the limit rather than read about it, open RV 10.106, eleven verses addressed to the Aśvins. It is a late hymn built almost entirely out of similes, piled up with no connecting narrative, many of them using compounds that occur nowhere else in Vedic.

Like the two pinions of a bird, connected, like two choice animals, ye have sought our worship. Bright as the fire the votary hath kindled, ye sacrifice in many a spot as roamers.

(RV 10.106.3, Griffith 1896) [1]

Like toiling bees ye bring to us your honey, as bees into the hide that opens downward.

(RV 10.106.10, Griffith 1896) [1]

Those are the parts that work. The middle of the hymn does not. Griffith, who translated all ten mandalas and rarely surrendered anything, wrote in his note on this hymn that he did not attempt “the hopeless task of translating stanzas 5, 6, 7, 8, in which nearly every word is a difficult riddle”, and filled the gap with H. H. Wilson’s renderings instead. [1] [12]

I came at this from the other end and found the same seam by accident. Working through this project’s copy of the Griffith text, I counted how many of its 9,535 verses contain the parenthetical supplements that a translator uses when the Sanskrit will not yield a sentence on its own. The answer is seven. Four of them are RV 10.106.5, 6, 7 and 8, sitting consecutively, in a prose style noticeably unlike Griffith’s verse. Of the remaining three, one is the notorious verse of RV 10.61 that Griffith left in Latin rather than English, a Victorian habit with its own peculiar history.

Aside. That observation is about a digital edition, not about the printed book; the files behind this site are a scrape of a public-domain text, and anomalies in them can be artefacts. The reason I trust it here is that Griffith’s own footnote says exactly what the data implies. It is a pleasant experience, finding a hole in a dataset and then finding that a translator apologised for it in 1897.

Modern scholarship has not closed the gap. Jamison and Brereton, whose 2014 translation is the standard, list several candidates for the hardest hymn in the Rigveda and say of this one that some of it appears to be written in unbreakable code. [13] The point to take from the hymn is not that the text is damaged. The metre is fine, the grammar is mostly fine, and the poet appears to have been showing off. What defeats us is a deliberately recherché vocabulary, invented or revived for one performance, by a poet whose audience knew the references and whose audience is dead.

How a dictionary admits defeat

The European encounter with these words produced a two-hundred-year argument about what to do when you cannot read a line, and the argument is more interesting than the stereotype of colonial philology suggests.

Max Müller’s six-volume edition, published between 1849 and 1874, printed the Rigveda together with the fourteenth-century commentary of Sāyaṇa, who worked at the Vijayanagara court. [14] Sāyaṇa’s bhāṣya has one striking property: it glosses everything. Every word gets a meaning, every verse gets a sense, and there is almost no admission of ignorance anywhere in it. H. H. Wilson, the first Boden Professor of Sanskrit at Oxford, took that as a reason to follow Sāyaṇa closely, on the straightforward argument that a learned brahmin in an unbroken tradition knew more than any outsider could. [12]

Rudolf von Roth took the opposite view: that Sāyaṇa stood some two and a half thousand years downstream of the hymns and was reconstructing their sense exactly as a modern scholar must, with fewer tools and more theological commitments. Roth’s position was that the Veda has to be interpreted from the Veda, by comparing a word’s own occurrences, and his dictionary with Otto von Böhtlingk, the seven-volume St Petersburg Sanskrit-Wörterbuch, built that principle into the reference work every later Indologist used. [16] The modern consensus sits closer to Roth than to Wilson, and uses Sāyaṇa the way it uses Yāska: as evidence about the tradition, and as a data point about what a very learned reader in 1360 thought a word meant. [15]

Read that column downwards and you are watching a discipline learn to say “I do not know” in writing. Grassmann’s contribution is easy to undervalue: by recording every occurrence of every word, his 1873 dictionary turned “this word is rare” from an impression into a number, and made the hapax a countable category rather than a feeling. [17] Lubotsky’s two-volume concordance of 1997 is the same instinct carried to completion. [2] Between the two of them sits Louis Renou, whose seventeen volumes of Études védiques et pāṇinéennes argue the hard passages hymn by hymn and are still where a specialist goes to find out how bad a given case really is. [23]

What does get recovered

None of this is a counsel of despair, because the record of recovery is real and it is instructive about method. Words have been pulled back from the dark, and almost always by evidence from outside the text.

graph TD
    A[Opaque Rigvedic word] --> B{More than one occurrence?}
    B -->|Yes| C[Compare contexts and metre]
    B -->|No| D[No internal distribution]
    C --> E{Iranian or IE cognate?}
    D --> E
    E -->|Yes| F[Sense narrowed by comparison]
    E -->|No| G{Used in later Vedic prose?}
    G -->|Yes| H[Sense inferred, drift likely]
    G -->|No| I{Material referent datable?}
    I -->|Yes| J[Archaeology fixes the object]
    I -->|No| K[Meaning uncertain]

Four kinds of outside evidence have done most of the work.

Comparison with Old Iranian is the oldest and strongest tool. The Avestan hymns are close enough to the Rigveda that a word attested in both can often be triangulated, and the shared religious vocabulary has clarified dozens of terms that internal evidence alone leaves vague.

Second, and more romantic, documents in other languages entirely. Indo-Aryan words embedded in Hittite and Mitanni material from the second millennium BCE, including a horse-training manual that counts chariot laps using numerals recognisably related to Vedic ones, give independent attestation of terms outside the Vedic tradition altogether. The Mitanni evidence is small in volume and enormous in value, because it is the only Indo-Aryan material of that age not transmitted by the recitation schools.

Third, archaeology, which fixes referents that no amount of philology can. When a word refers to an object, the object can be dug up. The Rigveda’s vocabulary for the light chariot and its spoked wheel is a worked example: the technology has a datable archaeological history, and that history constrains both the words and the date of the hymns that use them.

Fourth, botany and zoology, with mixed results. The identification of the soma plant is the famous case, and the famous warning: a century and a half of candidates, serious arguments for several of them, and no settled answer.

[!NOTE] Recoveries can also be reversed. F. B. J. Kuiper’s 1991 study proposed that several hundred Rigvedic words are of non-Indo-Aryan origin; Michael Witzel and others have argued that the defensible list is considerably shorter, and the disagreement turns on precisely the words that are too rare to analyse. [18] [19] A word reclaimed by one argument can be returned to the dark by the next.

Reading the brackets

Two conclusions, pulling in opposite directions, and both worth carrying.

The first is that the Rigveda’s opacity is not a defect in the transmission. It is the predictable cost of a transmission system optimised for sound. The hymns reached us because a tradition decided that the phonetic form of a mantra mattered absolutely and that paraphrase was worthless, and the same decision that preserved the pitch accent on a word like naicāśākha for three thousand years also meant that nobody was obliged to remember what it was. Had the tradition been willing to translate and modernise, as scribal cultures usually are, the text would be far more readable and far less authentic. We would understand more and have less.

The second is that the brackets in a modern translation are a kind of intellectual achievement. It took Vedic studies from Yāska to Jamison and Brereton, give or take twenty-five centuries, to arrive at a convention for printing uncertainty on the page rather than covering it with a confident gloss. The convention cuts against two very different instincts: the devotional one, which does not want to hear that anything was lost, and the modern technological one, which assumes that a large enough corpus and a clever enough model will eventually yield the meaning of anything. The second instinct is the one worth testing against this material. The Rigveda is now a thoroughly computational object, with a lemmatised dependency treebank and searchable corpora behind it. [8] Those tools are genuinely powerful for syntax and for statistics. They cannot fix a hapax, because a word that occurs once has no distribution to learn from, and no amount of annotation will tell you what the Kīkaṭas were doing with their cattle.

The Rigveda, read honestly, is an argument that meaning is not stored in texts. It is stored in the shared world that texts point at, and when that world goes, the text keeps its shape and loses its reference. Which is, when you think about it, a very Rigvedic observation: the poets of RV 10.71, a hymn about the making of language and the goddess Vāc, said something close to it about their own listeners.

One man hath ne’er seen Vāk, and yet he seeth: one man hath hearing but hath never heard her.

(RV 10.71.4, Griffith 1896) [1]

And in the great riddle hymn, more bluntly:

Upon what syllable of holy praise-song, as ‘twere their highest heaven, the Gods repose them. Who knows not this, what will he do with praise-song?

(RV 1.164.39, Griffith 1896) [1]

A practical suggestion to finish. Find RV 10.106 in two translations, one older and one current, and read them side by side with the verse numbers visible. Watch where the two diverge, where the brackets open, and where a confident Victorian sentence turns into a modern “meaning uncertain”. That divergence is not the failure of Vedic scholarship. It is the most honest page in it, and if you have ever wondered how we know what we claim to know about the ancient world, those four verses teach more than a whole shelf of summaries. For more on how the text’s own editors sorted and arranged what they could not always read, see the stratigraphy of the Rigveda; for the gods who survive as names with no stories attached, see Aja Ekapād and Ahi Budhnya.