AI Cracks Mystery of 4-Millennia-Old Code (Partially)

 Those of you who watched Terminator as a warning of the future may want to pack your bags: an artificial intelligence has just beaten the best human experts at a problem, one that some of them said was utterly impossible.  We recommend bringing some EMP weapons, and we hear that there's a nice spot for a city near the Earth's core.

The problem was that of the Indus script, writings left behind by the people of the Indus Valley four millennia ago - people thoughtless enough not to drop a dictionary into their ruins.  Only a few samples of the script survived, the longest only 27 characters long (not even a fifth of a Tweet!) which has posed significant problems in translation.

"They've tried to connect it to everything from Egyptian to Easter Island text, so they don't know what the hell it is"-kind-problems, with no success, and five years ago linguists started to claim it wasn't a language at all but a sequence of pretty pictures.  Which is about as much as a linguist can give up without going mute and wearing oven gloves.

Among the languages linked to the mysterious script are Chinese Lolo, Sumerian, Egyptian, Dravidian, Indo-Aryan, Old Slavic, even Easter Island — and, ultimately, no language at all. In 2004, linguist Steve Farmer published a paper asserting that the Indus script was nothing more than political and religious symbols.

Rajesh Rao of the University of Washington decided to apply computers to the conundrum, building a machine intelligence Markov model - a system for extracting patterns from data.  The artificial intelligence was loaded with data including ancient Sumerian, Sanskrit, human DNA, FORTRAN, bacterial proteins and an entirely artificial language.  We don't know about translation, but that sounds like a great way to summon a Cyber-Elder God to eat our faces.

The idea was to compare the level of pattern between human languages and other phenomena.  Adding the Indus script, Rao found it displays the same level of pattern as other languages - in other words, "Back to the grindstone, linguist boys!"  The program even output some grammatical rules of the language, but it still couldn't translate it.  Which, if Star Trek has taught us anything, means that somebody asked it to and it started smoking while lights frantically flashed before blowing up.

Posted by Luke McKinney.

http://blog.wired.com/wiredscience/2009/04/indusscript.html

Comments

[Burun Estetigi]

thanks

[ian]

So they ran a rank/order zipf on it and the slope on the log scale was diagonal. Is that what you're saying?