AI helped scholars recover a hymn to Babylon from scattered clay tablets

An ancient hymn celebrating Babylon, its gods and its people has re-emerged from surviving manuscripts. Anmar Fadhil and Enrique Jiménez published their edition online in March 2025. LMU credits its AI-supported electronic Babylonian Library, or eBL, with helping identify related copies.[1][2]

The edition documents 20 manuscripts, dating from the seventh to the second or first centuries BCE. It recovers roughly two-thirds, wholly or partly, of a poem that may originally have contained 250 lines. Among its subjects are the Euphrates bringing spring growth, the city’s religious life and the generosity of its inhabitants. School copies show that pupils encountered this work.[1]

Making damaged writing searchable

The crucial preparation is transliteration: recording cuneiform signs in a standardised textual form that a computer can search. The eBL dataset paper explains how specialists create these readings from tablet photographs, preserve information about damage and uncertainty, and attach each record to its museum object. A photograph and its reading therefore remain connected.[3]

This is detailed scholarly work before a search algorithm runs. The project’s workflow has a second expert review each transliteration and records later edits. Its 2024 data publication described approximately 25,000 transliterated tablets; that was the collection at the time, rather than a current inventory. The accompanying software makes the records available for computational research.[3]

The team’s 2024 methods paper describes n-gram matching: breaking sign sequences into short, overlapping groups and comparing those groups across records. Surviving copies of a work can contribute to a combined representation. That helps when a new fragment overlaps with several incomplete copies, each preserving a different part of the text.[4]

The result is a shortlist for expert checking. This method compares statistical patterns in surviving signs. Scholars establish whether proposed copies belong together and how their readings should be combined, retaining uncertainty where the evidence remains incomplete.[4]

That changes who can take part in reconstruction. Jiménez reports students identifying fragments through the online collection from their own homes. The practical gain is access to relevant evidence across collections. A researcher can spend more time considering a promising match and less time hoping that the next physical fragment requested from a museum happens to belong.[5]

Teaching AI to find the signs themselves

A June 2026 preprint describes a separate advance within the eBL project: computer vision that detects signs directly in photographs. The researchers applied their pipeline to 87,668 tablet fragments, producing nearly 2.9 million sign detections. These are proposed visual identifications, rather than millions of translated words. This later work was not the method responsible for the 2025 hymn edition.[6]

The pipeline first separates tablet views, then detects signs and groups them into lines. Damaged surfaces, uneven layouts and the absence of linguistic context still make the task difficult. Its contribution is to turn a large image collection into more structured material that scholars can inspect and computational methods can compare.[6]

Jiménez also wants to trace how texts refer to one another and travelled through scribal communities. Reconnecting their surviving copies makes that wider literary history easier to investigate. A readable work can reveal relationships that an isolated fragment leaves obscure.[5]

Report an error or suggest a correction