Israeli Experts use Artificial Intelligence to Translate Ancient Cuneiform

Ukrainian Union Conference

Israeli Experts use Artificial Intelligence to Translate Ancient Cuneiform

This is another important step toward preserving and disseminating the cultural heritage of Ancient Mesopotamia, scientists say.

Israel | Oleksandra Obrevko

Researchers from Tel Aviv University and Ariel University in Israel have developed an artificial intelligence model that can automatically translate cuneiform Akkadian text into English. This is another important step toward preserving and disseminating the cultural heritage of Ancient Mesopotamia, scientists say.

Assyriology specialists, who specialize in archaeological, historical, cultural, and linguistic studies of Assyria and the rest of ancient Mesopotamia, have spent many years trying to understand Akkadian texts written in cuneiform, one of the oldest known forms of writing. Cuneiform translates to "wedge-shaped" because in ancient times, people wrote in it with a reed stylus, making wedge-shaped marks on a clay tablet.

Now researchers from Tel Aviv and Ariel have developed an artificial intelligence model that will save all this effort. This model can automatically translate Akkadian text written in cuneiform into English.

This is reported by the educational project “Biblical Archaeology” with reference to The Jerusalem Post.

Assyria, named after the god Ashshur (the highest in the pantheon of Assyrian gods), was located on the Mesopotamian plain. In 721 B.C., the Assyrian army came from the north, captured the Northern Kingdom of Israel, and took ten tribes of Israel captive, after which they were lost to history.

Archaeologists have found hundreds of thousands of clay tablets in ancient Mesopotamia, written in cuneiform and dating back to 3400 B.C. However, there are many more tablets than can be easily translated by the limited number of experts who can read them.

Dr. Shai Gordin of Ariel University, Dr. Gai Gutherz and others from Tel Aviv University, and their colleagues published their findings in the PNAS Nexus journal on May 2, 2023, entitled "Translating Akkadian to English with neural machine translation” (Gai Gutherz, Shai Gordin, Luis Sáenz, Omer Levy, Jonathan Berant. PNAS Nexus, Volume 2, Issue 5, May 2023, pgad096).

When developing the new machine learning model, the researchers prepared two versions of automatic translation from Akkadian: one translates from Latinized inscriptions, and the other translates from cuneiform Unicode elements directly into English.

The first version, which uses Latin transliteration, gave more satisfactory results in this study, achieving a score of 37.47 in the Best Bilingual Evaluation Understudy 4 (BLEU4), which is a test of the level of correspondence between machine and human translation of the same text.

The program is most effective when translating sentences of 118 characters or less. In some sentences, it created "hallucinations," syntactically correct but inaccurate English results.

Dr. Gordin noted that in most cases, translation could be used as a primary text processing. The authors suggest that machine translation can be used as part of human-machine collaboration, where human scientists correct and improve the results of models.

Hundreds of thousands of clay tablets written in cuneiform document the political, social, economic, and scientific history of ancient Mesopotamia, the authors write, "but most of these documents remain untranslated and inaccessible due to their sheer number and the limited number of experts who can read them."

The scientists conclude that translation is a fundamental human activity that has a long scientific history since the emergence of writing. "It can be a complex process, as it usually requires not only expert knowledge of two different languages, but also different cultural environments. Digital tools that can help with translation are becoming more common every year, driven by advances in fields such as optical character recognition and machine translation. However, ancient languages still pose a serious challenge in this regard. Reading and understanding them requires knowledge of a long-dead language community, and the texts themselves can also be very fragmentary."

The original version of this story was posted on the Ukrainian Union Conference Ukrainian-language news site.