Development of a Curation System for Scribe Identification in Digital New Testament Manuscripts Using Deep Learning

2023 Fiscal Year Final Research Report

• Project/Area Number: 19K12714
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 90020:Library and information science, humanistic and social informatics-related
• Research Institution: Osaka University
• Principal Investigator: Miyake Maki 大阪大学, 大学院人文学研究科(言語文化学専攻), 准教授 (80448018)
• Project Period (FY): 2019-04-01 – 2024-03-31
• Keywords: IIIF / 深層学習

Outline of Final Research Achievements

This study leveraged IIIF-compliant digital images of New Testament majuscule manuscripts to analyze the characteristics of script shapes. In the Codex Vaticanus, differences in identification accuracy based on shape attributes of the script were observed, while in the Codex Sinaiticus, datasets were created based on character shape similarities. As a result of applying the deep learning method, One-Class Deep SVDD, an anomaly detection technique, overlaps in the differences between input-output and the distribution of anomaly scores were observed for the characters similar in shape to the reference characters. Additionally, it was suggested that elements other than shape might be involved in identification, as images with high anomaly scores often had cases where ink density or parts of adjacent characters were mixed in.

Free Research Field

Digital Humanities

Academic Significance and Societal Importance of the Research Achievements

人工知能分野の深層学習による写字識別のアプローチは、聖書学の伝統的正文批判研究や神学的解釈の固定概念からとは異なる観点を可能にし、客観的情報による新たな特徴の発見の可能性がある。高精細画像共有規格に準拠したIIIF対応の写本のデジタル画像を活用は、既存のデジタル資料の共有・再活用の実践的研究として意義があり、デジタル・ヒューマニティーズ分野の学際的研究の進展に貢献する。