The
Cuneiform ProjectScholars & collections
Help shape the reading workflow, assess difficult signs, and connect the work with images and scholarly editions.
A public-interest research initiative
The ancient world,
in its own words.
Written in clay. Preserved for millennia.
Help make its
stories accessible to everyone.
Our mission
A shared past.
A collective effort.
Receipts. Letters. Loans. Livestock counts. Cuneiform tablets preserve the details of lives lived thousands of years ago.
The Cuneiform Project is a public-interest research initiative developing AI-assisted cuneiform sign recognition. We bring machine learning and scholarship together to help specialists read ancient tablets while preserving the evidence behind each proposed sign.
Our goal is to help scholars read tablets at scale, connect their records, and uncover a richer picture of the ancient world: how people worked, traded, and lived.
Built for research. Grounded in evidence. Intended for everyone.
Cuneiform sign recognition
Every reading starts
with the evidence.
Our tools find possible signs in a photograph and suggest alternatives. A scholar can inspect the original image and make the call.
Photograph: British Museum, via CDLI. View source record
An inspectable reading
A suggestion.
A starting point.
Select a marked sign to see the model’s alternatives.
Recorded research output, not a live reading. Sign identities are provisional; they are not a translation.
Finding and recognizing signs is the first step. Reliable reading order, transliteration, and translation are still ahead.
About the research
Can the project translate a cuneiform tablet?
Not yet. The current research finds possible signs in a supplied tablet-face region and suggests visual sign identities for scholarly review. Reliable reading order, transliteration, and translation remain future work. The example above is a recorded research result, not a live translation service.
How can a proposed reading be checked?
Each proposal stays connected to its position in the original photograph and to ranked alternatives. A scholar can inspect that evidence before accepting or correcting a reading. Model scores are not established probabilities of correctness. The research note explains the measured results and their limitations.
Where does the example come from?
The demonstration uses tablet P335575, photographed by the British Museum and accessed through the Cuneiform Digital Library Initiative (CDLI). Sign identities are mapped through MZL identifiers to the ORACC sign list. These are source attributions, not claims of partnership. See the full sources and image credits.
A place to contribute
The next chapter
is a shared one.
This work brings together people who understand ancient writing, people who build new tools, and people who believe the past belongs to everyone.
Engineers & researchers
Advance sign detection, recognition, and the tools that keep every proposed reading connected to its evidence.
Partners & supporters
Help make careful annotation, research time, and computing resources available for the work ahead.