In the hope of uncovering new details about ancient life, researchers have developed a large language model that fills in the gaps in papyrus fragments.
A New Chatbot Wants to Unlock the Secrets in Tattered Ancient Greek Records
Business | By Joel Khalili | September 22, 2026, 5:30 AM
In the hope of uncovering new details about ancient life, researchers have developed a large language model that fills in the gaps in papyrus fragments.
The tattered remnants of ancient Greek records have long frustrated historians. Many of the papyrus scrolls that survived antiquity are riddled with gaps—missing words, broken lines, and eroded passages that have defied reconstruction for centuries. But a new AI system, developed by researchers at a leading university and published in a peer-reviewed journal, promises to fill in those blanks with unprecedented accuracy.
The tool is a large language model (LLM) trained specifically on ancient Greek texts. Unlike general-purpose models, which often hallucinate plausible-sounding but historically inaccurate words, this model was fine-tuned on a corpus of thousands of inscriptions and papyri, along with their known translations. It can predict missing characters and words in damaged passages, offering scholars a powerful new way to reconstruct fragmented documents.
How It Works
The system, which the researchers have nicknamed "Ithaca" (after the Homeric island), uses a deep neural network architecture similar to that of modern chatbots. When presented with a fragment containing gaps, it analyzes the surrounding context—both the visible text and the known conventions of ancient Greek—to generate a list of likely missing words or phrases, ranked by confidence.
Crucially, the model is not designed to replace human expertise. Instead, it acts as a collaborative tool, providing suggestions that historians can evaluate and refine. In tests, the model achieved a top-1 accuracy of 62 percent when predicting missing words, and a top-20 accuracy of 85 percent—significantly outperforming previous computational methods.
Why It Matters
The ability to reconstruct damaged texts could unlock new insights into ancient Greek society, from legal contracts and personal letters to literary works and religious inscriptions. Even small restorations can shift our understanding of historical events, trade networks, or everyday life.
"Every recovered word is a tiny window into the past," says one of the project's lead researchers. "If we can recover even a fraction of what was lost, we can ask new questions about how people lived, what they believed, and how they interacted."
2026 Context
The project is part of a broader trend in 2026: the application of AI to cultural heritage. Over the past year, similar models have been used to decipher cuneiform tablets, restore damaged medieval manuscripts, and even attribute disputed paintings. As computational power grows and digitized archives expand, researchers expect these tools to become standard in the humanities.
But the approach also raises questions. Who owns the reconstructions? How do we distinguish between a plausible guess and a historically valid restoration? The researchers emphasize that all AI-generated suggestions are clearly marked and subject to peer review.
For now, the team is making the model available to a select group of scholars, with plans for a broader release later this year. If successful, it could transform the way we read the past—one gap at a time.
via Wired AI