Algorithm seeks to decipher texts from a Bronze Age civilization

MADRID (EUROPA PRESS) -Florida Tech researchers are developing a machine learning algorithm to identify and digitize the enigmatic ancient writing of the Indus civilization.

Considered one of the first three civilizations in world history, along with Mesopotamia and Egypt, but geographically larger than those two, since it developed from the year 3300 BC in what is now Pakistan and India, the Indus civilization had with uniform weights and measures, skilled craftsmen, a multifaceted system of commerce, and more than 500 symbols and signs to communicate.

But scholars have been debating for decades whether these characters correspond to a language or are more similar to pictograms, as explained in a statement by Debasis Mitra, professor of computer science and lead author of the research.

The writings you are studying with your collaborators may be a series of symbols such as the equivalent of dollar signs and images of business transactions, or those symbols may be graphemes, individual letters or groups of letters that represent speech sounds.

The process uses an automated handwriting recognition (ASR) system to extract encoded grapheme sequences from a data set of more than a thousand photographs of Indus seals. Using two-stage artificial neural networks, ASR has achieved 88% success in detecting graphemes.

Still, the process has been challenging. Machine learning is often powered by inputting large amounts of data to basically train the system. In this case, however, there is not much data to enter. And what data there is can sometimes be “noisy” or distorted.

The team wants to continue refining their work until they create a system that allows archaeologists in the field to take a photograph of a text or symbol with a smartphone and send it to the database for digitization.