Google Translate has achieved a groundbreaking feat by using artificial intelligence (AI) to decode Akkadian, one of the world’s oldest languages. A team of computer scientists and historians, led by a Google software engineer and an Assyriologist from Ariel University, harnessed the power of Google Translate technology to develop an AI model capable of instantaneously translating the ancient characters found on cuneiform tablets.
Akkadian, the language of the Akkadian Empire that thrived in present-day Iraq from the 24th to 22nd century BCE, presents unique challenges for translation. Understanding its meaning has been compared to navigating without a North Star, as there are no descendant languages and a lack of cultural context. Cuneiform, a script distinguished by sharp, intersecting triangular symbols carved on clay tablets using the wedge-shaped end of a reed, was the writing system used by the Akkadians. But because there are so many tablets and so few experts who can understand them, a sizable chunk of these texts has remained untranslated and inaccessible.
The number of cuneiform texts in existence vastly outnumbers the number of Akkadian linguists. As a result, a vast amount of knowledge regarding this significant early civilisation—often referred to as the world’s first empire—has gone untapped. The growing number of tablets discovered by archaeologists is outpacing traditional linguistic efforts to decipher Akkadian writings. However, incorporating AI into the cuneiform interpretation process could transform it.
The AI model developed by the team focuses on two types of translation: cuneiform to English translation and cuneiform transliteration. The model’s translation quality, evaluated using the Best Bilingual Evaluation Understudy 4 (BLEU4) score, yielded remarkable results. Surpassing the team’s expectations, the model achieved scores of 36.52 and 37.47 for the two translation categories, providing high-quality translations. The BLEU4 score ranges from 0 to 100, with 70 representing the best possible result comparable to a highly experienced human translator.
In the past, computer-generated translations have been brittle and inaccurate because they have trouble understanding the subtleties of idioms and nonliteral words that violate established grammatical norms. Recent developments inh AI have explored the more complex facets of the language, such as the cuneiform translator. The AI translator still makes mistakes and occasionally experiences hallucinations despite its impressive accomplishments, which is typical of AI systems. Shorter lines and formulaic texts like administrative records translate well using the paradigm.
Additionally, it piques the curiosity of the researchers by capturing genre-specific traits during translation. The AI translator will eventually be trained on bigger sets of translations. It helps researchers by producing draft translations that can be revised and verified by human experts.