Create distinctive databases of ancient languages and using a computer vision model to accurately recognize and...

Area of Science:

• Digital Humanities
• Computational Linguistics
• Archaeology

Background:

• Cuneiform script is one of the oldest writing systems, originating in Mesopotamia.
• Deciphering ancient languages like cuneiform is a complex and time-consuming process.
• Training deep learning models for ancient scripts is challenging due to data acquisition difficulties.

Purpose of the Study:

• To develop a deep-learning-based sign detector for efficient identification and grouping of cuneiform tablets based on Hebrew content.
• To overcome the challenges of acquiring annotated training data for the Hebrew alphabet in cuneiform scripts.

Main Methods:

• Utilized pre-existing transliterations and a sign-by-sign Latin character representation to generate training data.
• Employed a supervised approach involving finding transliteration signs in tablet images and retraining a sign detector.
• Applied the Yolov8 object detection pretraining model for Hebrew character identification and cuneiform tablet categorization.

Main Results:

• The developed method effectively identifies Hebrew characters on cuneiform tablets.
• The sign detector's performance was enhanced, leading to improved alignment quality.
• Facilitated the categorization of cuneiform tablets based on their Hebrew content.

Conclusions:

• Deep learning offers a viable solution to accelerate the analysis of ancient cuneiform texts.
• The proposed method addresses data scarcity issues in training models for historical scripts.
• This research contributes to the digital humanities by enabling more efficient study of ancient languages and artifacts.