An as yet undeciphered script found on relics from the Indus valley constitutes a genuine written language, a new mathematical analysis suggests.
The finding is the latest chapter in a bitter dispute over the interpretation of 鈥淚ndus script鈥�. This is the name given to a collection of symbols found on artefacts from the Indus valley civilisation, which flourished in what is now eastern Pakistan and western India between 2500 and 1900 BC.
Advertisement
In 2002, a team of linguists and historians argued that the script did not represent language at all, but religious or political imagery.
Ordered or random?
From an analysis of the frequency and distribution of the script鈥檚 characters, the team concluded that it showed few of the hallmarks of language. Most of the inscriptions contain fewer than five characters, few of the characters repeat, and many of the symbols occur very infrequently.
The new analysis by computer scientist and his team at the University of Washington in Seattle comes to the opposite conclusion.
Rao鈥檚 team assessed the script samples using what is called 鈥渃onditional entropy鈥�. When aimed at language, this statistical technique comes up with a measure for the 鈥渙rderedness鈥� of words, letters or characters 鈥� from totally ordered to utterly random.
鈥淚f you look at strings that contain words, then you should see that for any particular word in the string there is going to be some amount of flexibility in choosing the next word, but they鈥檙e not randomly ordered,鈥� Rao says.
Which word next?
For instance, in English text, if you find the fragment 鈥淭he boy went to the鈥�, there is some flexibility in what follows. Nouns like 鈥減ark鈥� and 鈥渃ircus鈥� make sense, but a verb such as 鈥渆at鈥� does not.
Rao鈥檚 team applied this analysis to Indus script, Sanskrit, an ancient south Indian language called Old Tamil, and English. They also tested the conditional entropy of the Fortran computer programming language and non-languages, including DNA and protein sequences.
Indus script characters turned out to be about as randomly ordered as the other languages. Unsurprisingly, they proved less random than DNA or protein sequences and more random than the computer language, where unambiguity is essential.
Grammatical structure
鈥淣ow we can say, based on this evidence, that they probably were literate, so the big question becomes: Can you get at the underlying grammar?鈥� Rao says. He hopes to refine his team鈥檚 technique to determine the grammatical structure of Indus script and, potentially, the language family it belongs to.
鈥淚 think we are going to need more archival data, and if we are lucky enough we might stumble on a Rosetta Stone-like artefact,鈥� Rao says.
Rao鈥檚 paper has already drawn a strong response from the researchers who proposed that Indus script represents religious and political symbols, not language.
鈥淭here鈥檚 zero chance the Indus valley is literate. Zero,鈥� says , an independent scholar in Palo Alto, California who authored with two academics with the goading title 鈥淭he Collapse of the Indus Script Thesis: The myth of a literate Harappan civilization.鈥�
Simulated language
As well as comparing the conditional entropy of Indus script to that of known languages, they compared it with two simulated character sets 鈥� one totally random, one totally ordered.
Farmer and colleagues Michael Witzel of Harvard University and Richard Sproat of Oregon Health and Sciences University in Portland that the comparison with artificially created data sets is meaningless, as are the resulting conclusions. 鈥淎s they say: garbage in, garbage out,鈥� Witzel says.
Unlocking history
Farmer says that the debate over Indus script is more than academic chest thumping. If Indus script is not a language, a close analysis of its symbols could offer unique insight into the Indus Valley civilisation. Some symbols are more common in some geographical locations than others, and symbol usage seems to have changed over time.
鈥淵ou suddenly have a new key for unlocking how that civilisation functioned and what its history was like,鈥� he says.
, a linguist at the University of Wisconsin-Madison, says Rao鈥檚 paper is worth publishing, but time will tell if the technique sheds light on the nature of Indus script.
鈥淎t present they are lumping more than 700 years of writing into one data set,鈥� he says. 鈥淚 am actually going to be working with them on the revised analysis, and we will see how similar or different it is from the current results.鈥�
Journal reference: , (in press)
