They ran the biggest empire of their age, with a vast network of roads, granaries, warehouses and a complex system of government. Yet the Inca, founded in about AD1200 by Manco Capac, were unique for such a significant civilisation: they had no written language. This has been the conventional view of the Inca, whose dominions at their height covered almost all of the Andean region, from Colombia to Chile, until they were defeated in the Spanish conquest of 1532.

But a leading scholar of South American antiquity believes the Inca did have a form of non-verbal communication written in an encoded language similar to the binary code of today's computers. Gary Urton, professor of anthropology at Harvard University, has re-analysed the complicated knotted strings of the Inca - decorative objects called khipu - and found they contain a seven-bit binary code capable of conveying more than 1,500 separate units of information.

In the search for definitive proof of his discovery, which will be detailed in a book, Professor Urton believes he is close to finding the "Rosetta stone" of South America, a khipu story that was translated into Spanish more than 400 years ago.

"We need something like a Rosetta khipu and I'm optimistic that we will find one," said Professor Urton, referring to the basalt slab found at Rosetta, near Alexandria in Egypt, which allowed scholars to decipher a text written in Egyptian hieroglyphics from its demotic and Greek translations.

It has long been acknowledged that the khipu of the Inca were more than just decorative. In the 1920s, historians demonstrated that the knots on the strings of some khipu were arranged in such a way that they were a store of calculations, a textile version of an abacus.

Khipu can be immensely elaborate, composed of a main or primary cord to which are attached several pendant strings. Each pendant can have secondary or subsidiary strings which may in turn carry further subsidiary or tertiary strings, arranged like the branches of a tree. Khipu can be made of cotton or wool, cross-weaved or spun into strings. Different knots tied at various points along the strings give the khipu their distinctive appearance.

Professor Urton's study found there are, theoretically, seven points in the making of a khipu where the maker could make a simple choice between two possibilities, a seven-bit binary code. For instance, he or she could choose between weaving a string made of cotton or of wool, or they could weave in a "spin" or "ply" direction, or hang the pendant from the front of the primary string or from the back. In a strict seven-bit code this would give 128 permutations (two to the power of seven) but Professor Urton said because there were 24 possible colours that could be used in khipu construction, the actual permutations are 1,536 (or two to the power of six, multiplied by 24).

This could mean the code used by the makers allowed them to convey some 1,536 separate units of information, comparable to the estimated 1,000 to 1,500 Sumerian cuneiform signs, and double the number of signs in the hieroglyphs of the ancient Egyptians and the Maya of Central America.

If Professor Urton is right, it means the Inca not only invented a form of binary code more than 500 years before the invention of the computer, but they used it as part of the only three-dimensional written language. "They could have used it to represent a lot of information," he says. "Each element could have been a name, an identity or an activity as part of telling a story or a myth. It had considerable flexibility. I think a skilled khipu-keeper would have recognised the language. They would have looked and felt and used their store of knowledge in much the way we do when reading words."

There is also some anecdotal evidence that khipu were more than mere knots on a string used for storing calculations. The Spanish recorded capturing one Inca native trying to conceal a khipu which, he said, recorded everything done in his homeland "both the good and the evil". Unfortunately, in this as in many other encounters, the Spanish burnt the khipu and punished the native for having it, a typical response that did not engender an understanding of how the Inca used their khipu.

But Professor Urton said he had discovered a collection of 32 khipu in a burial site in northern Peru with Incan mummies dating from the time of the Spanish conquest. He hopes to find a khipu that can be matched in some way with a document written in Spanish, a khipu translation. He is working with documents from the same period, indicating that the Spanish worked closely with at least one khipu-keeper. "We have for the first time a set of khipu from a well-preserved and dated archaeological site, and documents that were being drawn up at the same time."

Without a "khipu Rosetta" it will be hard to convince the sceptics who insist that, at most, the knotted strings may be complicated mnemonic devices to help oral storytellers to remember their lines. If they are simple memory machines, khipu would not constitute a form of written language because they would have been understood only by their makers, or someone trained to recall the same story.

Professor Urton has little sympathy with this idea. "It is just not logical that they were making them for memory purposes," he said. "Tying a knot is simply a cue; it should have no information content in itself other than being a reminder." Khipu had layers of complexity that would be unnecessary if they were straightforward mnemonic devices, he said.

Translating the secrets of the ages

SUMERIAN CUNEIFORM

The world's first written language was created more than 5,000 years ago, based on pictograms, or simplified drawings representing actual objects or activities. The earliest cuneiform pictograms were etched into wet clay in vertical columns and, later, more symbolic signs were arranged in horizontal lines, much like modern writing. Cuneiform was adapted by several civilisations, such as the Akkadians, Babylonians and Assyrians, to write their own languages, and used for 3,000 years. Many of the clay tablets, and the occasional reed stylus used to etch cuneiform on them, have survived. Knowledge of cuneiform was lost until 1835 when a British Army officer, Henry Rawlinson, found inscriptions on a cliff at Behistun in Persia. They were identical texts written in three languages - Old Persian, Babylonian and Elamite - which allowed Rawlinson to make the first translation for many hundreds of years.

EGYPTIAN HIEROGLYPHICS

The original hieroglyphs, dating from about 5,000 years ago, were etched on stone and were elaborate and time-consuming to make, which meant they were reserved for buildings and royal tombs. A simplified version, called hieratic, was eventually developed for everyday bureaucracy, written on papyrus paper.

Later still, hieratic was replaced by demotic writing, the everyday language of Egypt, which appeared on the Rosetta stone with Greek and hieroglyphic script, allowing scholars to translate the original Egyptian writing.

MAYAN HIEROGLYPHICS

The Maya used about 800 individual signs or glyphs, paired in columns that read from left to right and top to bottom. The glyphs could be combined to form any word or concept in the Mayan language and inscriptions were carved in stone and wood on monuments or painted on paper, walls or pottery. Some glyphs were also painted as codices made of deer hide or bleached fig-tree paper covered by a thin layer of plaster and folded like an accordion. The complete deciphering of the Mayan writing is only 85 per cent complete, although it has been made easier with the help of computers.

Only highly trained Mayan scribes used and understood the glyphs, and they jealously guarded their knowledge in the belief that only they should act as intermediates between the gods and the common people.