Digital Hebrew Paleography: Script Types and Modes

Area of Science:

• Digital Humanities
• Computational Paleography
• Artificial Intelligence in Historical Studies

Background:

• Paleography, the study of ancient handwriting, is crucial for historical text analysis.
• Numerous medieval manuscripts remain unclassified due to the limitations of manual expert analysis.
• Automated tools are needed to efficiently classify large manuscript collections.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying medieval Hebrew manuscripts.
• To classify manuscripts into 14 distinct classes based on script style and graphical mode.
• To compare the performance of automated classification with human expert paleographers.

Main Methods:

• Utilized a deep learning methodology for script type classification.
• Experimented with various input image representations and network architectures.
• Implemented a hierarchical classification approach: first regional style, then graphical mode.
• Explored the use of soft labels and a 'squareness value' to refine graphical mode classification.

Main Results:

• Achieved highest accuracy with the hierarchical classification approach.
• Redefining graphical mode labels using the 'squareness value' significantly improved classification accuracy.
• The developed deep learning model demonstrated performance on par with human expert paleographers.

Conclusions:

• Deep learning offers an effective solution for automated classification of medieval Hebrew manuscripts.
• Hierarchical classification and refined labeling strategies enhance paleographic analysis accuracy.
• AI-powered tools can significantly aid researchers in processing and understanding historical textual data.