Deciphering ancient scripts has always been a monumental task, often demanding years of study, painstaking manual effort, and a deep understanding of historical context. Among the most enigmatic writing systems in human history is cuneiform, an ancient script that emerged in Mesopotamia over five millennia ago. Written by pressing a stylus into soft clay tablets, cuneiform consists of wedge-shaped characters that once recorded the daily lives, laws, trade, myths, and scholarly works of early civilizations like the Sumerians, Akkadians, Babylonians, and Assyrians. Yet, despite its profound historical significance, cuneiform remains only partially translated and understood, leaving vast troves of ancient knowledge locked behind undeciphered symbols.
Now, thanks to cutting-edge artificial intelligence (AI), scholars and researchers are making remarkable strides in cracking the code of this ancient writing system. A team of researchers from Cornell University and Tel Aviv University (TAU) have developed a revolutionary tool named ProtoSnap, offering an innovative solution to one of the oldest and most complex problems in the field of archaeology and ancient languages. This AI-powered system can automatically identify and reproduce cuneiform characters from photos of clay tablets, streamlining what was once an overwhelmingly difficult and manual process.
Cuneiform, alongside Egyptian hieroglyphs, stands as one of humanity’s earliest known forms of written communication. However, it’s far from a simple alphabet. The system incorporates over 1,000 distinct signs, each representing sounds, syllables, words, or concepts. To make matters more complex, the style and structure of these characters changed dramatically over millennia. Variations in handwriting, regional differences, and the evolution of scripts across different civilizations contribute to a chaotic patchwork of styles that challenge even the most seasoned experts.
“Even with the same character, its appearance changes across time,” explains Hadar Averbuch-Elor, assistant professor of computer science at Cornell Tech and in the Cornell Ann S. Bowers College of Computing and Information Science, who spearheaded the research. “It’s a very challenging problem to be able to automatically decipher what the character actually means.” This variability—what some might call the “handwriting” of ancient scribes—makes cuneiform transcription and translation extraordinarily difficult.
ProtoSnap addresses this problem with an ingenious AI solution. The researchers built a system capable of recognizing the variations in cuneiform characters and then “snapping” them into alignment with a generalized prototype of the character. In simple terms, the AI uses sophisticated image recognition techniques to identify minute differences and similarities between a specific character on an ancient tablet and an idealized, standardized version of that character.
The technology at the heart of ProtoSnap involves diffusion models, a type of generative AI frequently used for complex computer vision tasks such as image generation and enhancement. Diffusion models excel at understanding and generating images by analyzing pixel-level data. The team leveraged this capability to calculate the similarity between the pixels in an image of a character etched on a clay tablet and the corresponding pixels in a generalized prototype. Once the system determines the best match, it can align and snap the prototype into the exact contours of the character on the tablet. The result is a high-fidelity digital reconstruction of the character, tailored to the ancient scribe’s unique style but grounded in modern understanding.
Rachel Mikulinsky, a master’s student and co-first author from Tel Aviv University, will present their work, titled “ProtoSnap: Prototype Alignment for Cuneiform Signs,” at the prestigious International Conference on Learning Representations (ICLR) in April. Their presentation highlights not only the innovative technical approach behind ProtoSnap but also its broader implications for the study of ancient history and linguistics.
The significance of this breakthrough cannot be overstated. Today, it’s estimated that more than 500,000 cuneiform tablets are stored in museum collections worldwide. Yet only a small fraction of these ancient texts have been translated and published. Each translation can take weeks or even months, as experts painstakingly compare the shapes of characters to reference texts, accounting for damage, erosion, and stylistic variations. ProtoSnap offers a potential paradigm shift: it can automate and accelerate this process, saving scholars thousands of hours and making these ancient texts accessible at an unprecedented scale.
“There’s an endless amount of 2D scans of these cuneiforms,” said Averbuch-Elor, “but the amount of labeled data is very scarce.” In the realm of machine learning, labeled data—data that has been annotated by humans with explanations of what it represents—is essential for training AI systems. ProtoSnap overcomes this challenge by using its prototype alignment method to create high-quality, machine-readable versions of ancient characters, dramatically expanding the data available for training further AI models.
Once the ProtoSnap system has generated these clean, standardized versions of cuneiform signs, they can be fed into downstream AI models designed for optical character recognition (OCR). OCR is the process by which computers translate images of text into readable, editable machine text. When trained with data prepared by ProtoSnap, these OCR systems demonstrate significantly improved accuracy, even when faced with rare or highly variable characters. In contrast to previous AI efforts in deciphering cuneiform, which struggled with inconsistency and data scarcity, ProtoSnap provides a scalable, robust method for character recognition.
The broader implications of ProtoSnap extend far beyond simple transcription. By automating the process of copying cuneiform tablets, scholars can conduct large-scale analyses that were previously impossible. This means researchers could, for the first time, systematically compare writing styles across different regions, cities, or periods. Such comparisons could reveal insights about the diffusion of knowledge, trade relationships, political affiliations, or cultural shifts in ancient Mesopotamia and beyond.
Yoram Cohen, professor of archaeology at TAU and a co-author of the study, emphasizes the potential of ProtoSnap to transform the field. “At the base of our research is the aim to increase the ancient sources available to us by tenfold,” Cohen said. “This will allow us, for the first time, the manipulation of big data, leading to new measurable insights about ancient societies—their religion, economy, social and legal life.”
Indeed, one of the most exciting prospects raised by ProtoSnap is its ability to unlock “big data” analysis in ancient studies. In much the same way that data science has transformed modern industries—providing insights into consumer behavior, predicting trends, and informing policy—ProtoSnap could usher in a new era of quantitative archaeology. Researchers may soon be able to trace shifts in language, identify centers of administrative power, analyze legal documents for patterns of governance, or map out networks of trade and diplomacy based on the writings preserved in these ancient tablets.
Moreover, by dramatically reducing the time and labor involved in copying and translating tablets, ProtoSnap also democratizes access to this wealth of information. Traditionally, access to cuneiform tablets—either the originals or high-quality scans—has been limited to a small group of scholars. Now, with AI facilitating rapid transcription and translation, these ancient texts can be more easily digitized, published, and shared with the global academic community and the general public.
However, as with any technological advance, ProtoSnap also raises important ethical and practical considerations. Ensuring the accuracy and authenticity of AI-generated transcriptions is crucial, particularly as these digital copies become part of the scholarly record. Human oversight remains essential, and AI should be seen as a tool to aid—not replace—expert analysis. Likewise, as museums and institutions around the world hold vast collections of cuneiform tablets, questions about ownership, access, and digital rights will need to be addressed as these collections are digitized and studied at scale.
ProtoSnap represents a profound leap forward in the use of artificial intelligence to unlock the secrets of the ancient world. It bridges the gap between modern technology and ancient history, offering new ways to explore and understand humanity’s earliest written records. By making cuneiform transcription faster, more accurate, and more accessible, ProtoSnap promises to illuminate the rich cultural, economic, and spiritual life of Mesopotamia, bringing us closer to the voices of people who lived and wrote thousands of years ago.
In the end, deciphering cuneiform is not just about reading old inscriptions—it’s about reconnecting with the thoughts, ideas, and experiences of ancient civilizations. As scholars harness the power of AI, they open a new chapter in our ongoing quest to understand where we come from and how human society has evolved over millennia. ProtoSnap may be an AI tool, but its work echoes a deeply human pursuit: uncovering the stories and knowledge of our ancestors, long buried beneath time and clay.
More information: ProtoSnap: Prototype Alignment for Cuneiform Signs: tau-vailab.github.io/ProtoSnap/