An interesting technology advance is reported in New Scientist, and follows. What the researchers have done is quite clever, and probably sound.
They take an inscription by a known artisan of a known date, and store each letter in it. This gives them multiple letter ‘A’, for instance. They then create an ‘average’ image of these by superimposing all the ‘A’s on top of each other.
Then they take an unknown inscription, and create the same average image from the same letter. Then they compare how close it is. There’s probably a mathematical “index of similarity” function being used to decide this.
Apparently it works. They tried it out in a blindfold test on some known inscriptions, and it identified the artisan and the date correctly.
Computer reveals stone tablet ‘handwriting’ in a flash
18:00 02 July 2009 by Ewen Callaway
See a gallery of images showing the tablet-reading process
You might call it “CSI Ancient Greece”. A computer technique can tell the difference between ancient inscriptions created by different artisans, a feat that ordinarily consumes years of human scholarship.
“This is the first time anything like this had been done on a computer,” says Stephen Tracy, a Greek scholar and epigrapher at the Institute for Advanced Study in Princeton, New Jersey, who challenged a team of computer scientists to attribute 24 ancient Greek inscriptions to their rightful maker. “They knew nothing about inscriptions,” he says.
Tracy has spent his career making such attributions, which help scholars attach firmer dates to the tens of thousands of ancient Athenian and Attican stone inscriptions that have been found.
“Most inscriptions we find are very fragmentary,” Tracy says. “They are very difficult to date and, as is true of all archaeological artefacts, the better the date you can give to an artefact, the more it can tell you.”
Just as English handwriting morphed from ornate script filled with curvy flourishes to the utilitarian penmanship practiced today, Greek marble inscriptions evolved over the course of the civilisation.
“Lettering of the fifth century BC and lettering of the first century BC don’t look very much alike, and even a novice can tell them apart,” Tracy says.
But narrowing inscriptions to a window smaller than 100 years requires a better trained eye, not to mention far more time and effort; Tracy spent 15 years on his first book.
“One iota [a letter of the Greek alphabet] is pretty much like another, but I know one inscriber who makes an iota with a small little stroke at the top of the letter. I don’t know another cutter who does. That becomes, for him, like a signature,” says Tracy, who relies principally on the shape of individual letters to attribute authorship.
However, these signatures aren’t always apparent even after painstaking analysis, and attributions can vary among scholars, says Michail Panagopoulos , a computer scientist at the National Technical University of Athens, who led the project along with colleague Constantin Papaodysseus.
“I could show you two ‘A’s that look exactly the same, and I can tell you they are form different writers,” Panagopoulos says.
Panagopoulos’ team determined what different cutters meant each letter to look like by overlaying digital scans of the same letter in each individual inscription. They call this average a letter’s “platonic realisation”.
After performing this calculation for six Greek letters selected for their distinctness – Α, Ρ, Μ, Ν, Ο and Σ – across all 24 inscriptions, Panagopoulos’ team compared all the scripts that Tracy provided.
The researchers correctly attributed the inscriptions to six different cutters, who worked between 334 BC and 134 BC – a 100-per-cent success rate. “I was both surprised and encouraged,” Tracy says of their success.
“This is a very difficult problem,” agrees Lambert Schomaker,
a researcher at University of Groningen, Netherlands, who has developed computational methods to identify the handwriting of mediaeval monks, which is much easier to link to a writer compared with chisel marks on stone.Although Panagopoulos’ team correctly attributed all the inscriptions to their rightful chiseller, Schomaker worries that shadows could distort the digital photographs used in the analysis. Three-dimensional lasers scans of the inscriptions may offer more precision, he says.
Panagopoulos says his team is looking to use 3D images in the future.
The Greek computer scientists would also like to build a comprehensive database of digital inscriptions and attributions, so any newly discovered or analysed inscription could be quickly attributed and dated.
Journal references: Panagopoulos’ study – *IEEE Transactions on Pattern Analysis and Machine Intelligence*(DOI: 10.1109/TPAMI.2008.201); Tracy’s report – *American Journal of Archaeology* (Vol 113 (2009), No 1, p 99-102)