Silent letters from the past

Despite some high-profile success stories, deciphering ancient scripts is a long and arduous process in which even the most advanced technology is of only limited use. Andrew Robinson reports.

Fifty years ago, Michael Ventris announced one of the great intellectual achievements of the century on BBC radio: he had deciphered Linear B as an archaic form of Greek more than 500 years older than Homer. Europe's oldest readable writing was found at Knossos in Crete by Sir Arthur Evans. Ventris's achievement was one of a handful of great archaeological decipherments, beginning with Egyptian hieroglyphs in the 1820s and culminating with the Mayan glyphs in the last decades of the 20th century. But there are still eight or nine significant ancient languages that have defied decipherment,including Easter Island rongorongo , the seal stones of the Indus Valley civilisation in Pakistan/India, the hieroglyphic script of the kingdom of ancient Meroe in Sudan and the Etruscan alphabet of Italy.

Ventris summarised the three stages of the decipherment process masterfully. "An exhaustive analysis of the signs, words and contexts in all the available inscriptions, designed to extract every possible clue as to the spelling system, meaning and language structure; an experimental substitution of phonetic values to give possible words and inflections in a known or postulated language; and a decisive check , preferably with the aid of virgin material, to ensure that the apparent results are not due to fantasy, coincidence or circular reasoning."

Note that last phrase. Archaeological decipherment attracts both geniuses and cranks; and it is not always easy to tell the two apart. Sometimes, a successful decipherer of one script gets it into his head that he can decipher others and becomes a crank. Undeciphered scripts, the rongorongo scholar Jacques Guy unceremoniously declares, are "powerful kook attractors".

What are the minimum conditions for any degree of decipherment to be feasible? According to Ventris, "the material should be large enough for the analysis to yield usable results, and (in the case of an unreadable script without bilinguals or proper names)" - ie a script that lacks a Rosetta stone - "the concealed language should be related to one that we already know".

In all the successful decipherments would-be decipherers have had masses of text to work with - eventually, if not initially. Alas, that is not true for any of the undeciphered scripts. The largest corpus belongs to Etruscan, with some 13,000 inscriptions (many of them fragmentary or graffiti) mainly of a funerary nature; the smallest is that of the notorious Phaistos disc of Crete, with a mere 242 text characters making up 45 different signs. Since its signs write a language of which we know nothing, there is absolutely no hope of deciphering the disc unless more of the script turns up in excavations - though this has not stopped dozens of people, including a few classical scholars, from claiming to have deciphered the disc following its discovery in 1908.

In each successful decipherment, too, the language underlying the script was assumed to be related to a known language: for example, Coptic (for Egyptian hieroglyphs), Greek (Linear B), and various Mayan languages of Central America (Mayan glyphs). With the ancient scripts still to be deciphered, the best hope lies in identifying a modern language that is likely, on firm historical and cultural grounds, to be related to the script's concealed language: a licence for scholars to attempt to reconstruct the ancestor of the modern language.

There is much uncertainty in such reconstructions, given the magnitude of change in languages over the centuries and millennia. The best-known one is a proto-Indo-European language, spoken perhaps 6,000-8,000 years ago, reconstructed from structural similarities between words in Sanskrit, Greek, Latin and other European languages. Reconstruction of the language of Easter Island's rongorongo is theoretically possible given the strong likelihood that it is closely related to well-known Polynesian languages. But efforts to reconstruct the Etruscan language from Indo-European languages, especially Latin and Greek, have proved fruitless. Etruscan seems to be an "isolate", unrelated to any Indo-European language.

To return to the signs, independent of any consideration of their associated language, there are numerous possibilities for analysis of the kind mentioned by Ventris. Two elements of an unknown script usually yield their secrets without too much effort.

The first is the direction of writing: from left to right or from right to left, from top to bottom or from bottom to top. Clues to the direction include the position of unfilled space in the text, the way in which characters sometimes crowd (on the left or on the right), and the direction in which pictographic signs face. But there are certain scripts that are written boustrophedon (from the Greek for "as the ox turns", when ploughing), in other words first from left to right (say), then from right to left, then again from left to right, and so on. There are even reverse-boustrophedon scripts, in which the writer turned the original document through 180 degrees at the end of each line; rongorongo is an example of this.

The second element is the system of counting. Numerals frequently stand out graphically from the rest of the text, especially if they are used for calculations (which helpfully suggests that the non-numerical signs next to the numerals are likely to stand for counted objects or people). Easily visible numerals are a particular feature of the Linear B and Mayan scripts and, among the undeciphered scripts, of the proto-Elamite script (of ancient Iran); and the numerical system is obvious in the Etruscan script, the Linear A script (of Crete and the Aegean) and the Zapotec and Isthmian scripts (of Mexico), fairly clear in the Indus Valley script, but seems to be largely absent from the Meroitic script and rongorongo , and not at all evident in the Phaistos disc.

More challenging than direction of writing or numerals is the analysis of the sign system as a whole. Suppose you were unfamiliar with the Roman alphabet. If you were to take a typical chapter of an ordinary novel printed in English, it would be a fairly straightforward matter, by careful study and comparison of the thousands of characters in the text, to work out that they could be classified into a set of signs: 26 lower-case letters and a similar number of upper-case signs (though you might wonder whether letters with ascenders such as b, d, f, h, k should be classified with the lower-case or the upper-case letters) plus sundry other signs, mainly punctuation marks and numerals. Imagine that the same text is handwritten. Immediately, the task of isolating the basic signs is far harder, because the letters are joined up and different writers write the same letter in different ways, also differently from its printed equivalent, and not always distinctly.

If the signs and sign variants of an undeciphered script can be correctly classified - and it is a big "if" - scholars can get a fair idea of whether the script is an alphabetic/ consonantal script (20-40 signs), a syllabary (40-90 signs), or a logosyllabic script with hundreds of syllables and logograms ("whole-word" signs like £+ or %) - without having any idea of the phonetic values of the signs. For example, the 20th-century decipherers of Ugaritic, a cuneiform script from Syria (ancient Ugarit), quickly realised that it had only 30 signs and therefore could not be a logosyllabic script like Babylonian cuneiform. The script of the Phaistos disc, with 45-plus signs, is likely to be syllabic.

A numbered sign list can be made, and each undeciphered inscription can be written in terms of a sequence of numbers instead of the usual graphic symbols, and thereby classified by computer in a concordance. Then one can ask the computer to calculate the relative sign frequencies, or to list all the inscriptions in which a particular sign occurs; and, further, all the inscriptions in which a particular combination of signs occurs. If one suspects this combination of representing, say, a certain word or proper name, one can then analyse in exactly which contexts (at the beginning of inscriptions, in the middle of words, next to which other signs) the combination occurs, within every inscription in a corpus.

Although such frequency analysis has been applied to the Linear A, Meroitic and Indus script corpuses, computers have made little impact in archaeological decipherment. One reason is the difficulty of discriminating between signs and sign variants, which is still a matter of human judgement; another is the great graphical complexity of many scripts, which does not lend itself to the black-and-white, discrete nature of numerical classification. Yet another reason, more general, is that there is not really enough text for computerised statistical techniques to prove decisive. On the whole, successful decipherment has turned out to require a synthesis of logic and intuition based on wide linguistic, archaeological, historical and cultural knowledge that computers do not possess.

This is especially relevant to the question of what constitutes "sense" in the results of a decipherment. How do you judge whether the meaning is right at the pivotal moment when you conduct "an experimental substitution of phonetic values to give possible words and inflections in a known or postulated language", as Ventris said?

Those who approach decipherment expecting sensational revelations of great battles and the fall of civilisations or grand philosophical discussions and astrological prognostications, or even visitations from extraterrestrials - as one rongorongo "decipherer" has claimed - are likely to find their expectations confirmed in the mysterious signs of an undeciphered script, even if they have to invent an underlying language with no plausible relationship to any known language (the correctness of which can naturally never be proved or disproved).

But as John Chadwick, Ventris's academic collaborator on Linear B and a wartime Bletchley Park codebreaker, said: "The simplest, most mundane and least surprising explanation of any inscription is likely to be the correct one."

Thus it was reasonable to suppose that the undeciphered Linear B tablets would contain abbreviated bureaucratic accounts, not paeans to the gods or epic poetry such as Homer's, for at least three simple reasons. First, the tablets plainly contained many numerals and pictograms of mundane objects (vessels, animals, chariots and so on); second, they were scratched on a cheap and relatively impermanent medium, clay, without much care for aesthetics, unlike the beautiful contemporary Minoan seals carved on gemstones; and lastly, they were discovered in what was obviously a palace archive, like the much larger palace archives of clay tablets found in Mesopotamia written in cuneiform.

Given the surprising and fascinating epigraphic revelations of life among the ancient Maya since the 1970s, maybe Chadwick was being typically over-cautious about what we may yet find in the undeciphered scripts. But then he had been exposed to dozens of highly imaginative "decipherments" of the alluringly enigmatic Phaistos disc. While there is no hope of deciphering this unique object, so long as it remains unique, the other undeciphered scripts may still have surprises in store.

Andrew Robinson is the author of Lost Languages : The Enigma of the World's Undeciphered Scripts , McGraw-Hill, £25.99, and The Man Who Deciphered Linear B : The Story of Michael Ventris , Thames and Hudson, £12.95.