Tamil Nadu put up $1 million for the Indus script. Nobody can claim it.

In January 2025, at a conference in Chennai marking a hundred years since the Indus civilisation was announced to the world, the chief minister of Tamil Nadu offered a million dollars to anyone who could read the Indus script. Twenty months later the money is still there.

That is not for want of trying. Since the 1920s, more than a hundred decipherments have been published by archaeologists, epigraphists, linguists, engineers, hobbyists and cryptographers. Several were announced as final. None has been accepted. An unclaimed prize is a blunt instrument, but as a reading of where the field actually stands, it beats any press release.

What there actually is to read

The numbers first, because the numbers are the whole problem.

Roughly four thousand inscribed objects have been found, and a 2025 count puts it nearer five thousand depending on what you are willing to call an inscription. The average one is about five signs long. The longest sequence anyone has found on a single flat surface, a seal catalogued M-314, runs to seventeen.

Seventeen signs. That is the longest sentence the Indus civilisation has left us, if it is a sentence at all.

You will sometimes see twenty six quoted instead. That figure comes from a pair of small moulded three-sided prisms, and it is reached by adding up the signs on all three faces. Whether that makes one text or three is exactly the sort of question nobody can settle without being able to read it.

The sign list is contested too. Iravatham Mahadevan's 1977 concordance settled on 417 distinct signs, and that number became the standard because nearly everything computational since has been built on his corpus. Bryan Wells counts closer to 694. The disagreement is not carelessness. Without a reading there is no way to tell a separate sign from a stylistic variant of another sign, so the inventory is a judgement call. What the counts do agree on is the shape of the thing: about 67 signs carry four fifths of all usage, trailed by a long tail of signs that turn up once or twice and never again.

And there is nothing else. No papyrus, no palm leaf, no birch bark, no wooden tablets. Whatever the Harappans may have written on anything organic, the soil of the Indus plain has not kept it. The largest single inscription in existence is the signboard at Dholavira in Kutch: ten signs, each about 37 cm high, cut from gypsum and inlaid along a wooden board roughly three metres long. It stood over the north gate of the citadel until it fell flat on its face, and the wood rotted away leaving the letters lying where they landed. R.S. Bisht's team found it in 1991. Ten signs is the biggest thing we have.

Why a century has not been enough

The reason the script is hard is easy to state. There is no bilingual.

Champollion had the Rosetta Stone, the same decree written twice, once in Egyptian and once in Greek. Michael Ventris had thousands of Linear B accounting tablets and, eventually, the realisation that the language underneath was an early form of Greek, which he could check against a language he already knew. The Indus script has neither advantage. No bilingual text has ever been found. The language it may encode is unidentified and may have no living descendants. The script stops around 1900 BCE and nothing obviously grows out of it.

So anyone attempting it is solving for three unknowns at once: what the signs are, what language they write, and what the texts say. Any one of those would be hard. Together they will let you produce almost any answer you like, which is precisely why there are over a hundred of them.

The other prize, the one that says there is nothing to read

In 2004, Steve Farmer, Richard Sproat and Michael Witzel published a paper with a title that did not invite discussion: "The Collapse of the Indus-Script Thesis: The Myth of a Literate Harappan Civilization". Their argument is that the Indus symbols are not writing at all.

The case rests on the same numbers as everything else. Real scripts, they say, do not behave like this. The inscriptions are impossibly brief and stay brief across seven centuries of use, which is not what happens when a writing system settles in. Rare signs multiply over time instead of dying out, again the wrong direction. And the texts lack the untidy, random-looking repetition that spoken language always leaves behind in writing. What the symbols do resemble, they argue, is a non-linguistic emblem system of the kind that carried religious, political and administrative weight across the ancient Near East without ever encoding speech. Their comparison, and they meant it seriously, was road and airport signage.

Then they did something unusual for a journal article. They put up $10,000, funded by an anonymous donor and good for Farmer's lifetime, on one condition:

Turn up just one Indus inscription that contains at least 50 symbols distributed in the outwardly random-appearing ways typical of true scripts contemporary to the Indus system, and we will (1) declare our model to be falsified and (2) will write a check to the discoverer or discoverers for $10,000.

The inscription has to come from a known Indus site and be accepted as genuine by the field. The offer has stood for more than twenty years. It is also unclaimed.

So there are two prizes outstanding on the same pile of evidence. One is for reading it. The other is for proving there is nothing there to read.

The fight that went statistical

In 2009 a team led by Rajesh Rao at the University of Washington published a paper in Science measuring the conditional entropy of Indus sign sequences, which is roughly a measure of how predictable the next sign is given the one before it. On that test, the Indus corpus sat with natural languages rather than with DNA, protein sequences or Fortran code.

Sproat's reply, in Computational Linguistics, called the work "useless" and "trivially and demonstrably wrong", and produced artificial sequences that scored similarly without being language at all. Rao answered that this was a strawman, that he had never claimed entropy on its own proved anything, and that the result was one of several properties pointing the same way. The exchange ran on until 2015 across two journals. Neither side gave an inch.

It is worth being clear about what that settled, which is nothing. It produced no reading. What it did establish is that the statistical evidence is compatible with both positions, a result that suits nobody.

The best argument for a Dravidian reading, and the man who doubted it

The most influential decipherment programme is Asko Parpola's, and its most famous piece of evidence is a fish.

The fish sign is common in the corpus. In Old Tamil, min means fish, and it also means star, from a root meaning to glitter. If the Harappans wrote by rebus, using a picture of one thing to stand for a word that merely sounds like it, then a drawn fish could be read as star. On seal H-9, seven vertical strokes stand beside a fish, and elu min, seven stars, is the Old Tamil name for Ursa Major. It is an elegant argument, and you can see why it has held people for fifty years.

The complication is who doubted it. Mahadevan, whose concordance the whole field runs on, and who himself believed the language was Dravidian, wrote in a 1994 review of Parpola's book that he had misgivings about this particular reading. His settled position later was that the fish signs really do depict fish, but that they mean something other than stars, perhaps through a different Dravidian fish word or some extended sense we have not recovered. The Indologist Harry Falk has put a blunter objection: the Sumerians, the Akkadians and the Chinese all drew a star as a star. Why would the Harappans not?

Why none of this is a quiet academic matter

Whether the Indus language was Dravidian or Indo-Aryan is not a neutral technical question in India. A Dravidian reading puts the ancestors of the south Indian languages at the centre of the subcontinent's first cities. An Indo-Aryan or Sanskritic reading ties the Harappans to the Vedic tradition. Both get recruited by people who want them to settle something about who belongs where, and the pressure shows.

In 2000, N.S. Rajaram announced that a Harappan seal depicted a horse. That would have mattered a great deal, because horses are conspicuously scarce in Harappan remains and central to Vedic society. Writing in Frontline that October, Witzel and Farmer showed that the horse was a computer-manipulated image of a broken unicorn seal, the fracture smoothed over and the animal's outline altered. They published a follow-up the next month. The episode is worth remembering less as a point scored against one side than as a demonstration of what the stakes do to the evidence.

What we know anyway

Here is the part that gets lost in the decipherment fights. Almost everything worth knowing about the Indus civilisation was worked out without reading a word of it.

We know the seals were stamps rather than amulets, because pressed into wet clay they left a sealing that marked a bundle of goods. We know the goods travelled, because Indus seals and sealings turn up at Ur, at Kish and at Susa, in a Mesopotamia that traded with a place it called Meluhha. We know the writing runs right to left in the impression about nine times out of ten, worked out from the way signs get cramped at one end when the carver ran out of room. We know a one-horned animal appears on the great majority of seals, and that a jar-shaped sign is the commonest of the lot.

We know the cities had gridded streets, standardised fired bricks, covered drains, and a weight system so consistent that the little stone cubes from sites hundreds of kilometres apart still match each other. We know there are no palaces anyone can point to and no obvious royal tombs. We know it ran for roughly seven centuries at its height and then came apart.

All of that is archaeology. None of it needed the script. What the script would add is the one thing digging cannot give you: the Harappans in their own words, saying whatever it was they thought worth carving onto a stamp.

What a million dollars would have to buy

Mahadevan is the right person to end on. He joined the Indian Administrative Service in 1953 and stayed in it for twenty seven years. He took a Nehru Fellowship in 1970 to work on the script, published the Concordance in 1977, and kept at it until he died in Chennai in 2018, aged 88. He did more than anyone alive or dead to make the corpus usable by other people. He never claimed to have deciphered it.

The prize was announced alongside a two crore rupee chair endowed in his name and research funding for the Indus Research Centre at the Roja Muthiah Research Library in Chennai. It followed a study by K. Rajan and R. Sivanantham which digitised more than 15,000 graffiti-bearing potsherds from 140 sites across Tamil Nadu and reported that over 90 per cent of the marks have parallels among Indus signs.

That study deserves the same scepticism as everything else here, and it has had it. The Tamil Nadu megalithic pottery is roughly a thousand years younger than the mature Harappan script, and Falk's objection stands unanswered: marks scratched on pots were practical marks on pots, not necessarily a system for communicating anything. Visual resemblance between simple scratches is cheap.

Which puts the prize back where it started. To collect it, somebody has to produce a reading that satisfies people who have spent entire careers disagreeing with each other, working from a corpus whose longest text is seventeen signs, in a language nobody has identified, with no bilingual anywhere. The money looks safe. That is the finding.