Is Sanskrit the mother of all languages?

“Sanskrit is the mother of all languages.”

You have almost certainly met the sentence. It usually arrives with a citation attached, and the citation is usually Sir William Jones, the East India Company judge who told Europe that Sanskrit was kin to Greek and Latin. The name is what gives the claim its weight.

Jones wrote no such thing.

What he did write, in a paper read to the Asiatic Society in Calcutta on 2 February 1786, says close to the opposite. It is the most quoted passage in the history of the subject, it has been in print since 1788, and it is almost always quoted in halves.

Here is the whole of it:

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

The half everyone knows is the first half. More perfect than the Greek. More copious than the Latin. That half is a compliment, and it has been carried for two hundred and forty years as proof that Sanskrit sits at the head of the human family of languages.

The half that settles the argument is the last clause. Some common source, which, perhaps, no longer exists.

The parent is the thing that is gone. Sanskrit is one of the children.

What Jones proposed, and the priest who got there first

Jones was not the first European to notice that Sanskrit resembled Latin and Greek. The Florentine merchant Filippo Sassetti, writing home from Goa in 1585, had already listed the pairs: deva and dio, sarpa and serpe, sapta and sette, nava and nove. Marcus Zuerius van Boxhorn proposed a common ancestor for Dutch, German, Latin, Greek and Persian in 1647, and insisted on excluding loanwords and comparing grammar instead of vocabulary, which is the correct method arrived at a century and a half early.

The most instructive predecessor is a French Jesuit named Gaston-Laurent Cœurdoux, who died at Pondicherry in 1779. In a memoir sent to Paris in 1767, or 1768 by some accounts, Cœurdoux laid out the Sanskrit, Latin and Greek correspondences carefully and in detail. He had the same evidence Jones had. He reached a different conclusion.

Cœurdoux explained the resemblance as borrowing. Neighbouring nations, long in contact, trading words back and forth.

That is the whole difference between them, and it is the hinge of this entire subject. Two languages can look alike because one gave words to the other, or because both inherited them from a third. The first is contact. The second is descent. They leave different evidence and they mean entirely different things, and almost every confused claim about Sanskrit comes from mistaking one for the other.

The historian Thomas Trautmann, whose Languages and Nations is the standard account of this period, sets the two men side by side:

“He accounts for the similarity of the three languages not by co-descent from a single ancestor language, as in Jones, but by mutual borrowing among languages long neighboring one another, though originally distinct… Jones gives an explanation of language similarity through co-descent, positing a movement from original unity to difference… It is the explanation of Jones that became the foundation of comparative philology.”

Jones’s contribution was never the observation. It was the mechanism.

And the mechanism is a family tree, which means it produces sisters. Sanskrit stands beside Avestan, Greek, Latin, Gothic and Old Church Slavonic. Above all of them sits a language nobody ever wrote down and nobody has ever read, reconstructed backwards from its children and marked with an asterisk in every textbook to say so. Linguists call it Proto-Indo-European. It has no literature, no speakers, no agreed date and no homeland everyone accepts. It exists the way a missing ancestor exists in a genealogy: as the only thing that explains the cousins.

Where the claim actually came from

So if Jones did not say it, who did?

The English sentence “Sanskrit is the mother of all languages” has no traceable author. It cannot be sourced to a scholar, a text, or a date. It is a folk accretion, repeated with confidence by people who believe they are quoting somebody.

The position underneath it is another matter. That has a name and a year.

In 1808, at Heidelberg, Friedrich Schlegel published Über die Sprache und Weisheit der Indier, and stated it plainly. Comparison of the languages, he wrote, yields the result that the Indic language is the older, and the others later and derived from it. The philologist W. P. Lehmann, introducing the passage in his standard reader of nineteenth-century linguistics, describes Schlegel’s position precisely: he derives the European languages from Sanskrit on the ground of its greater antiquity, without positing any intermediate stage.

That is the claim in its original form. It is not Indian in origin and it is not ancient. It was made in German, in Germany, twenty-two years after Jones, by a scholar of the Romantic movement who had learned Sanskrit in Paris.

For a while it was live and serious. Franz Bopp, whose 1816 comparison of the conjugational systems founded the discipline proper, would not commit either way. His formula was “all the languages which stem from Sanskrit or from a mother language in common with it.” The or is his. He did not know, and he said so.

Then the field spent seventy years taking it apart.

August Schleicher, in his Compendium of 1861 and 1862, was the first to attempt a systematic reconstruction of the parent language itself, and the first to draw the family tree. Once a parent could be reconstructed, Sanskrit had a place inside the tree instead of at the top of it.

The decisive blow came in the 1870s, and it was narrower and stranger than anything so far.

The comparativists the comparativists worked out what is now called the Law of the Palatals. The Indo-Iranian languages have a set of palatal consonants where their European cousins have plain velars, and the puzzle was why the split fell where it did. The answer was that the palatals had arisen before front vowels, one of which Sanskrit subsequently merged away into its plain a. Which meant that Sanskrit’s famously economical vowel system was not the ancestral one at all. It was an innovation. The messier Greek arrangement, with its separate e, a and o, was the conservative one, and it is Greek that preserves the parent’s shape at this point.

Sanskrit had been taken for the oldest partly because it looked the most orderly. Here, its orderliness turned out to be the change.

Edwin Bryant’s survey of the whole period, in The Quest for the Origins of Vedic Culture, gives the chapter its title without ceremony: “Indo-European Comparative Linguistics: The Dethronement of Sanskrit.”

Note what dethroned it. The comparative method that Sanskrit itself had made possible, using Sanskrit’s own evidence.

An Irishman in Tirunelveli

While that was going on in Europe, the question moved south in India, into the hands of a man who had come for something else entirely.

Robert Caldwell was born on 7 May 1814 at Clady in County Tyrone, to poor Ulster Scots Presbyterian parents. He arrived in Madras on 8 January 1838, four months short of his twenty-fourth birthday, sent by the London Missionary Society. He was posted to the Tirunelveli district at the southern tip of the peninsula and he stayed there for over fifty years. He married Eliza Mault in 1844. They had seven children. He was made Bishop of Tirunelveli in 1877. He died on 28 August 1891 and was buried at Idayankudi, in the district he had spent his life in.

He learned Tamil because he needed to preach in it.

What he produced instead, in 1856, was A Comparative Grammar of the Dravidian or South-Indian Family of Languages, and it demonstrated something the scholarship of his day had assumed away: that Tamil, Telugu, Kannada, Malayalam and their relatives do not descend from Sanskrit. They constitute a separate family with its own history and its own reconstructible parent.

He gave the family the name it still carries.

What survived, and what did not

Caldwell’s core finding is the foundation of an entire discipline. Bhadriraju Krishnamurti, whose The Dravidian Languages (Cambridge, 2003) is the standard modern reference, works from it as settled ground and counts over twenty-six Dravidian languages in four genetic subgroups.

Not everything in the 1856 book survived, and a piece that reported only the vindication would be doing exactly the selective quoting it is complaining about.

So: the record, itemised.

Caldwell argued in his second edition, at pages 91 and 92, that the oldest Dravidian word preserved anywhere in writing is the word for peacock in the Hebrew of Kings and Chronicles, in the list of goods brought back in Solomon’s ships. The Hebrew tuki, he proposed, is the old Tamil and Malayalam tokei. Walter Eugene Clark went after this directly in the American Journal of Semitic Languages and Literatures in 1920, citing Caldwell by page, and concluded that the comparison “is of no historical value.” His objection was chronological rather than phonetic. He doubted the peacock had that name in Tamil in the tenth century BCE. And the argument is not closed even now: David Shulman, in Tamil: A Biography (Harvard, 2016), accepts both the peacock and Caldwell’s second candidate, the aloes-wood.

Caldwell also proposed that Greek oryza, rice, came from Tamil arisi. Krishnamurti corrects him by name: the comparison is with a reconstructed Proto-Dravidian form, “and not with Ta. arisi… as proposed by Caldwell.”

He affiliated Dravidian to a group he called Scythian, meaning roughly what is now called Ural-Altaic. That affiliation was never established. It was also never refuted. Krishnamurti’s summary of the whole genre of such proposals is that “none of these hypotheses has been proved beyond reasonable doubt,” and Robert Austerlitz, reviewing them in the early 1970s, called most of them unprofessional and unsystematic. Kamil Zvelebil, in his Dravidian Linguistics of 1990, thought the Ural-Altaic connection the most promising of a bad lot. That is where it sits. Unproved, unburied.

And one of his speculations turned out to be right and is still taught. Caldwell noticed that place names in Pliny and Ptolemy tend to end in -our or -oura. Krishnamurti accepts it: the ending is the Dravidian ur, town.

One etymology corrected, one contested for a century and a half, one quietly standing, and the core holding firm underneath all of it.

Tamil Nadu put him on Marina Beach. The statue went up in 1968, alongside one of Ilango Adigal, for the world Tamil conference held in Madras that January. India Post issued a commemorative stamp on 7 May 2010. The state government observed his bicentenary as an official function in 2014, beginning at his tomb in Idayankudi.

The finding that Tamil is not Sanskrit’s daughter is not a Western correction imposed on an Indian belief. It is honoured in bronze, by the state whose language it concerns.

What Tamil already had

There is a version of this story in which Caldwell hands Tamil its dignity. That version is wrong, and the corpus says so.

By the time Caldwell landed at Madras, Tamil had behind it the Sangam poetry, the body of work produced by the academies of poets that tradition places in the ancient Tamil country. Zvelebil, working from the best available edition, counts 2,381 poems. Of these, 102 are anonymous. The remaining 2,279 are ascribed to 473 poets known by name or by epithet, though Zvelebil notes that thirty-five of those names are pseudonyms coined from an image in the poem itself, which would bring the count of real named poets nearer 438.

Among them are women. Zvelebil names Auvaiyar, “the Old Lady,” and Kavarpendu, “the Female Guard,” and Kamakkanniyar. Auvaiyar is not a token presence in the margins either. She is one of the sixteen most prolific poets in the whole corpus, with fifty-nine surviving poems.

The dating is contested and should stay that way, but Zvelebil’s conclusion is stated in his own words in The Smile of Murugan, weighed against linguistic, epigraphic, archaeological, numismatic and historical evidence: “the earliest corpus of Tamil literature may be dated between 100 B.C. and 250 A.D.”

The poems are organised into the eight anthologies, the Ettuttokai, and the ten longer pieces, the Pattuppattu. And they run on a theory of their own making. Sangam poetry divides the world into akam, the interior, which is love, and puram, the exterior, which is war and kingship and public conduct. Under akam sits the tinai system, in which five landscapes, the mountain and the forest and the farmland and the seashore and the wasteland, each carry a fixed emotional register, so that naming a place is already naming a state of the heart.

Tamil did not merely exist early. It had theorised itself. That history runs forward into the bhakti saints and the temple south, and it is told in the Tamil stream.

The inversion

Now turn the question around, because the strongest evidence against Sanskrit as universal mother is inside Sanskrit’s own oldest text.

The Rigveda contains words borrowed from Dravidian.

Krishnamurti states it on page six of his book: Rigvedic Sanskrit “already had over a dozen lexical items borrowed from Dravidian,” and he lists ulukhala, mortar, and kunda, pit, and khala, threshing floor, and kana, one-eyed, and mayura, peacock.

Count, though, is exactly where the field argues. M. B. Emeneau went through T. Burrow’s proposed borrowings with a hard filter and threw most of them out, remarking that not all of Burrow’s suggestions would survive Burrow’s own principles. Twelve came through as very definite. Of those twelve, only three actually occur in the Rigveda, though Emeneau elsewhere adds a fourth, bala, strength, from a Dravidian root meaning to be strong. Southworth took twenty. Nikita Gurov, in a conference paper, put it as high as eighty words across 146 hymns. The figures run across an order of magnitude, which usually means a field is arguing about method rather than about evidence.

So take the hostile witness.

Michael Witzel has argued more forcefully than anyone that Dravidian influence on the oldest layers of the Rigveda is close to nothing. His verdict at the end of his 1999 monograph on the substrate languages of Old Indo-Aryan is deflationary and precise:

“There is very little Dravidian, but there are about 300 words of the Indus substrate… At best, one can speak of a few very isolated cases which have been taken over into the RV; clearly this indicates an adstrate rather than a substrate.” For the earliest books he allows only a hedged shortlist, ukha-chid, hip-breaking, at 4.19.9, phalgu, minute or weak, at 4.5.14, and ani, lynch pin, at 5.43.8, and then says openly that whether even this is enough to place Dravidian speakers in the Punjab at that date “may remain in the balance.”

He is the last person with an interest in conceding the general point. He concedes it anyway.

Dravidian loanwords, he writes, “suddenly appear” in the middle and late layers of the text. And again, later: “Rgvedic loans from Drav. are visible.”

He names them with verse numbers. Kavasha, straddle-legged, used as a personal name, at 7.18.12. Kula, slope or bank, at 8.47.11. And, he allows, perhaps kunda, vessel, at 8.17.13. From the late layers, khala, the threshing floor, at 10.48.7, and katuka, pungent, at 10.85.34.

The dispute is about how many and how early. It has never been about whether.

There is a bird in the middle of all this that deserves its own paragraph. Mayura, peacock, appears in the Rigveda in compounds and in a feminine form at 3.45.1, 1.191.14 and 8.1.25. Krishnamurti lists it among the Dravidian loans, with the Tamil mayil behind it. Witzel takes it away and gives it to Munda, the Austroasiatic family, deriving it from a proto-form meaning “crier,” and in the same breath records the Dravidian counter-argument, which is that in Dravidian the word for the peacock’s feather reconstructs to a level older than the word for the bird. And it is the same bird Caldwell thought he had found in Solomon’s ships, argued over since 1875 and unsettled now.

One word. Three language families. Still open.

Then the part that should stop anyone who wants a tidy story. Those three hundred substrate words are roughly four percent of the Rigveda’s priestly vocabulary, and Witzel cannot confidently say what language they come from. He calls the source Para-Munda. Krishnamurti rejects the label and says so bluntly: it would have been better, he writes, to say that we do not know the true source of those three hundred early borrowings.

The two men agree about the words and disagree about everything else. Which means the oldest Sanskrit we possess is in conversation with a language nobody has been able to name.

That is the real picture of the Rigvedic world, and it is stranger and better than the slogan. If you want the archaeological and genetic layer underneath it, that is in the Aryan migration debate and in the Indus Valley and the question of Hindu origins, and it is worth keeping the two questions apart. Genes are not languages.

The traffic runs the other way too

The borrowing was never one-directional, and it did not stop with vocabulary.

Krishnamurti reports that the Rigvedic language carries over 380 loanwords from non-Aryan sources, of which 88 contain retroflex consonants, the sounds made with the tongue-tip curled back that are among the most recognisable features of Indian speech and are largely absent from Sanskrit’s European sisters. He notes too that the Rigveda uses the gerund, which Avestan does not, with the same function it has in Dravidian, and uses iti as a quotative marker to close reported speech, again as Dravidian does. Following Kuiper, he takes these as marks of a substratum rather than of simple word-borrowing. Structure travels harder than vocabulary, and it travels only under sustained contact.

Suniti Kumar Chatterji, the Calcutta linguist who wrote The Origin and Development of the Bengali Language in 1926 and turned directly to the question in “Non-Aryan Elements in Indo-Aryan” ten years later, had made the case for exactly this decades before it was fashionable. Krishnamurti’s own conclusion is that the structural evidence points to the absorption of a Dravidian substratum into Indo-Aryan “from the earliest stages of contact, which gradually affected its grammatical structure over three millennia.”

Three thousand years of traffic, in both directions, and none of it descent.

Which brings the argument to its sharpest test, from the other side.

The Tolkappiyam is the oldest surviving Tamil grammar. It runs to something near sixteen hundred aphorisms in three books, though the exact figure moves depending on whose division of the verses an edition follows: Zvelebil reports 1,595 on Ilampuranar’s division and 1,611 on Naccinarkkiniyar’s, and 1,571 from an editor who stripped out what he judged to be interpolations. Its date is genuinely open. Zvelebil places the nucleus in the second or first century BCE and no earlier than 150 BCE, with final redaction no earlier than the fifth century CE. Iravatham Mahadevan argued from the pulli, the dot used to mark a bare consonant, that it can be no earlier than the second century CE. Vaiyapuri Pillai placed it in the fifth or sixth. Takahashi put the oldest layers in the first or second century CE. Herman Tieken pushed it to the ninth and has been widely criticised for it.

Here is the thing worth noticing. The founding grammar of the language that owes Sanskrit no descent knew Sanskrit grammar well.

Zvelebil examined one aphorism in the second book, Collatikaram 419, and found it followed Patanjali’s classification of compounds so closely that it seemed “almost a translation” of the Sanskrit. Elsewhere he connects a rule in the first book to Panini. A. C. Burnell had earlier argued that the whole work was modelled on the pre-Paninian Aindra school of grammarians.

And the plainest evidence is a single word. The Tolkappiyam’s own term for the discipline of grammar is ilakkanam, which is the Sanskrit lakshana naturalised into Tamil. The word for grammar is a loanword. Zvelebil uses it to date the book.

Borrowing is not descent. English is full of French and owes France nothing genealogically. Tamil took its word for grammar from Sanskrit and remains no relation. The Rigveda took its threshing floor from Dravidian and remains Indo-European to the root.

What is actually true about Sanskrit

Sanskrit’s importance to comparative linguistics is real, and it is the fact the false claim is a distortion of.

Sanskrit is conservative. It held on to more of the parent language’s shape, in its verb roots and its case endings, than most of its sisters did. When nineteenth-century philologists went looking for the outline of Proto-Indo-European, Sanskrit is where the outline showed through most clearly. It made the reconstruction possible. That is a genuine and unusual distinction, and it is the opposite of being everyone’s mother. It is being the child who kept the family photographs.

Then there is Panini.

The Ashtadhyayi, the Eight Chapters, describes the Sanskrit language in roughly four thousand rules. The precise number is a trap, and George Cardona, the leading modern authority on Panini, explains why: the count usually cited comes from the Kashikavritti recension, which has 3,983 sutras and which Cardona notes “represents an inflated text.” Panini’s own date is unsettled and has been moving. The older view, argued on linguistic grounds by Paul Thieme and Hartmut Scharfe, puts him in the sixth or fifth century BCE. More recent work by Oskar von Hinüber and Harry Falk, reasoning from numismatic evidence, presses toward the middle of the fourth. Cardona’s own careful statement sets only the outer bound: the evidence “hardly allows one to date Panini later than the early to mid fourth century B.C.”

What the Ashtadhyayi does is harder to overstate than to state. It is not a list of observations. It is a machine: rules governed by rules about rules, ordered so that a general provision yields to a specific one, deriving the forms of the language by transformation from underlying elements, using recursion.

The comparison with modern computing goes back a long way. In March 1967 a one-page letter appeared in Communications of the ACM in which P. Z. Ingerman proposed that the notation then called Backus Normal Form, used to specify programming languages, be renamed the Panini-Backus Form, on the ground that Panini had invented a notation equivalent in power to Backus’s some twenty-three centuries earlier. The suggestion was never adopted.

The comparison, examined properly, says something better than the compliment it was meant as. In 2012 the computer scientist Gerald Penn and the linguist Paul Kiparsky tested it. They found the resemblance to modern rewriting systems to be “striking but cosmetic,” and the Paninian formalism to be very much more powerful than a programming grammar, capable of generating string languages that lie outside the classes computational linguistics can comfortably handle. That is not a compliment on its own. A formalism of that power is a liability, because it can describe almost anything, including nonsense. Their real finding is the next step: Panini’s grammar, they write, “judiciously avoided the potential pitfalls of this unconstrained formalism.”

The achievement was never the raw power. It was the restraint. John Kadvany, who has traced a genuine parallel between Panini’s use of auxiliary markers and the rewriting systems Emil Post devised in the 1920s, states the limit just as plainly: Panini’s grammar as a whole is not a computing language and not a Post rewrite system.

And the reach is real. Sanskrit vocabulary sits inside Tamil, Telugu, Kannada and Malayalam as tatsama, words taken over unchanged, and tadbhava, words assimilated to the borrowing language’s sound patterns. It sits inside Thai, Khmer, Javanese and Malay. Very few languages in the world have a history resembling it. But that history is the history of a prestige language being taken up, and prestige has conditions attached: who was permitted to learn Sanskrit, which Indian traditions refused it outright, and how it nonetheless crossed Asia to Java are taken up separately in who was allowed to learn Sanskrit.

What people mean when they say it now

Which brings us to the thing many readers will have arrived with.

There is a claim, circulated for decades and repeated in classrooms, that NASA has determined Sanskrit to be the best language for computers, or that its future supercomputers will run on Sanskrit.

It descends from a single article. Rick Briggs, “Knowledge Representation in Sanskrit and Artificial Intelligence,” published in AI Magazine in the spring of 1985. Briggs worked at the Research Institute for Advanced Computer Science, a joint venture between the Universities Space Research Association and NASA, based at NASA Ames. He was not a NASA employee.

What he argued is narrow, technical and genuinely interesting. Taking what he calls Shastric Sanskrit, the language of the grammatical and philosophical treatises, he showed that when its analysts decomposed a sentence into karaka relations, the roles that link an action to the things taking part in it, they produced the same set of triples that a modern semantic network produces when it is decomposed into nodes, arcs and labels. His flourish at the end is that it is tempting to think of the Sanskrit grammarians as computer scientists without the hardware.

Here is what the article does not contain. It does not mention supercomputers. It does not mention programming. It makes no claim that Sanskrit is more suitable for computing than any other language. Search the full text and the word NASA appears exactly once, in the author’s affiliation line at the top, and nowhere else in the piece.

A companion claim, that Forbes magazine declared Sanskrit the best language for software in July 1987, appears to point to an article that does not exist. A request for a copy on a mailing list of Sanskrit scholars produced nothing.

The tradition Briggs was writing about is extraordinary, which is why the paper was worth writing. The claim built on top of it is somebody’s invention, and it does exactly the work the slogan does. Both take a real and specific achievement and inflate it into a universal one, and the inflation always runs in the same direction.

Two careful men, flattened

Look at what happened to the two scholars in this story.

Jones wrote a long, cautious, conditional sentence proposing a lost common ancestor, and it was cut in half and turned into a slogan asserting Sanskrit’s primacy. Caldwell wrote a technical comparative grammar demonstrating an independent language family, and it was taken up after his death by political movements he did not live to see and would not have recognised.

Neither man was misquoted by accident. In both cases what was removed was the qualification, the conditional, the part where the scholar said what he did not know. In both cases the removal served somebody’s argument.

The position itself is still held, and it has a name in the scholarly literature. Michael Witzel, cataloguing the views that dispute the mainstream account of Indo-Aryan origins, separates three that are regularly muddled together. One holds that the Indo-Aryans originated in the Punjab. A second, the Out of India theory, holds that the Iranians migrated out of it. The third and most intense, in his words, “has all languages of the world derived from Sanskrit,” and he calls it the devabhasa school, from devabhasa, the speech of the gods. It is, he adds, “mostly, but not solely, restricted to traditional Pandits.”

It is the third that this piece has been about. The first two are different arguments, with different evidence, and they deserve to be met on their own terms rather than folded into this one.

The correction, in every direction, is the same correction. Read the whole sentence.

What the tradition actually claims

There is one more thing, and skipping it would make everything above a category error dressed as a refutation.

When the tradition calls Sanskrit daivi vak, the speech of the gods, it is not proposing a family tree.

The oldest statement about speech anywhere in the corpus is the hymn at Rigveda 10.125, in which Vak, speech herself, addresses the listener in the first person and describes her own reach. She is a goddess. That is a claim about status, and about what language is, and it has nothing whatever to say about which language descends from which. Her later form is taken up in Who is Saraswati.

The Veda is held to be apaurusheya, without human author. It was not composed. It was seen. The Mimamsa school, whose business is precisely the interpretation of Vedic language, holds that shabda, sound or word, is eternal, and that the relation between a word and its meaning is no human convention. The Nyaya school disagreed, and held that relation to be samketa, agreement, established by usage. That is a real and ancient argument about the nature of language itself, and it is set out further in what is Mimamsa. Bhartrihari, in the Vakyapadiya, moved the question onto different ground altogether and identified the ground of reality itself as shabda-brahman, the absolute as word.

These are metaphysical claims. They can be argued with on metaphysical grounds. They are not claims that Sanskrit is the historical ancestor of French, and no amount of comparative philology touches them, because comparative philology is not asking their question.

Most people who repeat the slogan mean something close to the first claim. They have been handed the second.

And the second costs the first something. A claim about revelation is not the kind of thing linguistics can test, and it stands or falls on other grounds entirely. A claim about descent is testable. It has been tested. It failed. Attaching the failed one to the untested one does not strengthen anything. It hands anyone who wants to dismiss the whole thing an easy target, and it makes an old and serious position about the nature of language look like a bad guess about history.

Sanskrit does not need to be everyone’s mother. Nothing in the Vedas, nothing in Mimamsa, nothing in Bhartrihari requires it.

Back to Calcutta

Go back to the room in 1786.

A judge who had learned Sanskrit in order to read Hindu law stands up and says that three languages resemble each other too closely for accident, and that the resemblance points backward to a source which, he suspects, is gone. He does not claim to know what it was. He does not claim to know where. The word he chooses is perhaps.

The paper went into Asiatick Researches in 1788 and it has been in print ever since. Anybody could have read the whole sentence at any point in the two hundred and forty years since.

It has been saying the same thing the entire time.

The sentence is not the problem. The half is.