An interdisciplinary team of computer science and history researchers used artificial intelligence (AI) to translate the world's oldest language, Akkadian.
Led by a Google software engineer and an Assyriologist from Ariel University, the team has employed the same technology that powers Google Translate to create an AI model capable of instantly deciphering the ancient glyphs found in cuneiform tablets.
Akkadian, the language of the Akkadian Empire that thrived in present-day Iraq during the 24th to 22nd centuries BCE, poses unique challenges for translation.
With no descendant languages and a dearth of cultural context, comprehending its meaning is akin to traveling without a North Star. The Akkadian writing system employed cuneiform, characterized by sharp, intersecting triangular figures, which were inscribed on clay tablets using the wedge-shaped end of a reed.
"Hundreds of thousands of clay tablets inscribed in the cuneiform script document the political, social, economic, and scientific history of ancient Mesopotamia," the research team emphasizes. "Yet, most of these documents remain untranslated and inaccessible due to their sheer number and the limited quantity of experts able to read them."
The enormity of the existing cuneiform texts far exceeds the limited number of linguists proficient in Akkadian. Consequently, a vast repository of knowledge about this significant early civilization, often considered the world's first empire, remains untapped.
Linguistic efforts to translate Akkadian texts struggle to keep pace with the increasing number of tablets being excavated by archaeologists. However, the integration of AI into the cuneiform interpretation process holds the potential to transform this landscape.
The AI model developed by the team excels in two types of translation: cuneiform to English and cuneiform transliteration (rewriting phonetically). The model's translation quality, measured by the Best Bilingual Evaluation Understudy 4 (BLEU4) score, yielded impressive results.
Scoring 36.52 and 37.47 for the two translation types respectively, the model surpassed the team's expectations, providing high-quality translations. The BLEU4 score ranges from 0 to 100, with 70 being the highest achievable score for a highly skilled human translator.
Traditionally, computer-generated translations have been brittle and unreliable, unable to capture the full richness of idioms and nonliteral language that elude formal grammatical rules. However, recent advancements in AI, such as the cuneiform translator, have delved into the more nuanced aspects of language.
Despite its remarkable achievements, the cuneiform AI translator still produces errors and occasional "hallucinations," a common occurrence in AI systems.
The AI model demonstrates the highest accuracy when translating shorter sentences and formulaic texts like administrative records. It also reproduces genre-specific nuances during translation, a discovery that intrigued the researchers. The AI will be trained on larger samples of translations in the future.
Currently, the AI translator aids researchers by generating preliminary translations that can be refined and verified by human experts.
No comments:
Post a Comment