The Indus Script

Summary

A Bronze Age civilisation that rivalled Egypt and Mesopotamia left behind roughly four thousand inscribed objects, mostly seals, carrying several hundred distinct signs.

For a century, people have tried to read them. In 2025 the chief minister of Tamil Nadu offered a million dollars to anyone who could.

But there is a prior question, and it is the one that makes this case different from every other undeciphered script in this archive. It is not settled that the Indus signs are writing at all.

What is documented

The corpus. Around four thousand inscribed objects. They are overwhelmingly seals: small carved stamps, typically with a line of signs across the top and an animal figure beneath. They also appear on tablets, on pottery, and on a small number of other objects.

The signs. Estimates of the sign inventory range from roughly four hundred to six hundred distinct characters, depending on how a scholar counts variants.

The length, which is the whole problem. The average inscription is about five signs long. The longest known is on the order of a couple of dozen.

There is no Indus text. There are Indus labels.

No bilingual. Nothing equivalent to the Rosetta Stone has ever been found, and nothing that could serve the function has ever been found.

Where the signs appear. Overwhelmingly on objects of commerce. Seals are for stamping goods. Rajesh Rao, a computer scientist at the University of Washington who has worked on the script for over a decade, has noted that a great many attempted decipherments read spiritual or religious meaning into the signs, and that this sits awkwardly with the fact that the signs appear mostly on the ancient equivalent of shipping labels.

The prize. In January 2025, M. K. Stalin, chief minister of Tamil Nadu, announced a prize of one million dollars for a decipherment of the script to the satisfaction of archaeological experts. It followed a study by K. Rajan and R. Sivananthan that digitised some fifteen thousand graffiti-marked potsherds from a hundred and forty sites in Tamil Nadu and compared them against around four thousand Indus examples, reporting morphological parallels.

The politics, which are real and dangerous. The identity of the Indus language is a live political question in India. One position holds that it was an ancestor of the Dravidian languages, which would imply that Dravidian languages were widespread across the subcontinent before the arrival of Indo-Aryan speakers. Another holds that Sanskrit and its relatives originated within the Indus Valley and spread outward from it.

Respected scholars hold both positions, and the registry takes no view on either. What it records, because it is a documented fact about the case, is that researchers working on the Indus script have received death threats.

Leading explanations

The prior question: is it writing?

Farmer, Sproat and Witzel (2004) argued that it is not. Their paper, The Collapse of the Indus-Script Thesis, made three principal arguments, and each of them is an argument from a real feature of the corpus.

The inscriptions are too short. Around five signs. Full writing systems produce long texts, and the Indus Valley, across seven centuries and a very large area, produced none.

There are too many rare signs, and the problem gets worse. In a maturing writing system, the sign inventory stabilises and rare signs are shed. In the Indus corpus, the number of rare signs increases across the seven hundred years of the Mature Harappan period. That is the wrong direction.

The sign repetition is wrong. Real writing produces the random-looking repetitions that language produces: the same sign twice in a row, the same short sequence recurring. The Indus corpus does not show this in the expected way.

Their proposal was that the signs are a non-linguistic system: emblems, ownership marks, commodity and tax notations, religious or clan symbols. Meaningful, structured, and not encoding speech.

They were attacked for it. They also offered a standing challenge: produce a single Indus inscription of more than fifty signs. Nobody has.

Rao and colleagues (2009) replied with statistics. Writing in Science, they measured the conditional entropy of the Indus corpus, which is a measure of how predictable each sign is given the one before it, and reported that it falls closer to the values of natural languages than to those of various non-linguistic sign systems.

Sproat replied to the reply. He argued that the method lacks discriminative power: applied to known non-linguistic systems, such as Mesopotamian deity symbols, it produces similar results. The exchange continued in Computational Linguistics in 2010 and in Language in 2014, and Sproat has since restated the central objection plainly: the presence of structure in a symbol system is not, by itself, evidence that the system encodes language.

He is right about that, and it is the crux. Structure is necessary. It is not sufficient.

Where it stands. Nobody has demonstrated that the Indus signs encode a language, and nobody has demonstrated that they do not. The statistical work established that the signs are not random, which was worth establishing and which nobody seriously doubted. What kind of non-random they are remains open.

Current status

Unsolved, and unusually so.

Four thousand objects. Several hundred signs. An average inscription of five characters. No bilingual. No long text. A century of attempts, a million-dollar prize, and an unresolved argument about whether there is anything there to read.

The registry keeps this record as the hardest case in its category, and as the one where the honest answer is the least satisfying: we do not know what the Indus signs mean, and we do not know whether they mean anything in the sense that words mean things.

Last reviewed: July 2026. Records are provisional. Where the evidence changes, the entry changes. Found an error? Tell us.