Who organized the Tbilisi conference, and who funded it?

It was organized by the Academy of Digital Humanities – Georgia and funded by Georgia’s Shota Rustaveli National Science Foundation, chaired by Prof. Manana Tandashvili of Goethe University Frankfurt with Dr. Mariam Kamarauli as project lead.

Is the manuscript-decoding AI available to the public?

No. It is research software developed within the ERC-funded DeLiCaTe project at the University of Hamburg, used by philologists studying Georgian and Armenian palimpsests rather than released as a public tool.

Tbilisi Conference Applies AI to Decode Ancient Georgian Manuscripts

Q: What is a palimpsest?

A palimpsest is a manuscript whose original text was scraped off centuries ago so the parchment could be reused, leaving only faint traces of the erased writing beneath later layers of text.

Q: How does AI help reveal the erased text?

Researchers pair multispectral imaging, which captures manuscripts across light wavelengths invisible to the eye, with generative AI models trained to reconstruct faded or overlapping text layers into clearer, more readable results than raw scans.

An international conference held in Tbilisi, Georgia, from September 1 to 6 brought together linguists and computer scientists to discuss how artificial intelligence can help recover text lost from some of the country’s oldest manuscripts, according to organizers at the Academy of Digital Humanities – Georgia.

Titled “Artificial Intelligence and Modern Methods and Technologies for Digital Processing of Georgian Manuscripts,” the gathering centered on palimpsests — manuscripts whose original text was scraped off centuries ago so the parchment could be reused, leaving only faint traces of the erased writing beneath later layers. Reading those traces has traditionally required years of manual philological work.

Much of the featured research builds on DeLiCaTe, a five-year project at the University of Hamburg funded by the European Research Council and led by philologist Jost Gippert, who also gave a keynote at the Tbilisi conference. Researchers on the project pair multispectral imaging — photographing manuscripts across light wavelengths invisible to the eye — with generative AI models trained to reconstruct faded or overlapping text layers, producing readings the team describes as clearer than raw multispectral scans. The technique targets Georgian and Armenian palimpsests specifically, including material tied to the wider history of literacy in the Caucasus.

The Tbilisi conference itself was chaired by Prof. Manana Tandashvili of Goethe University Frankfurt, with Dr. Mariam Kamarauli serving as project lead, and was funded by Georgia’s Shota Rustaveli National Science Foundation. Additional keynote speakers included Lado Chanturia and Zviad Gabisonia. Organizers kept the event small, capping attendance at 20 participants chosen from abstracts submitted earlier this year — making it more of a working session for active researchers than a public showcase.

Sessions also covered other AI challenges specific to Georgian, including automated hate-speech detection, comparisons of machine and human translation quality, and computational analysis of literary style — the kind of work where Georgian researchers increasingly compete internationally, as seen recently when Georgian students won four medals at the first European AI Olympiad.

The manuscript research joins a small but growing list of AI projects aimed at Georgia’s cultural and natural heritage, including a system built earlier this year to identify the country’s native cattle breeds from photographs — part of a broader pattern of adapting AI tools built for data-rich languages and domains to smaller, tradition-rich ones with limited digital records.

Frequently asked questions