What question did this study set out to answer?

The project aims to design a digital library framework for managing contested historical records, preserving accessibility and context.

May 9, 2026Open Access

Beyond Archives: Designing a Segmentation and ATR Workflow for Mid-Twentieth-Century Typescripts in a Contested-Memory Case Study

Key Points

The project aims to design a digital library framework for managing contested historical records, preserving accessibility and context.
Developed a workflow for layout segmentation and automatic text recognition (ATR) of mid-twentieth-century Italian typescripts.
Utilized PAGE/ALTO outputs and TEI-oriented structuring for enhanced data organization.
Emphasized controlled correction and paradata documentation to ensure data integrity and reusability.
Demonstrated effective integration of institutional records and oral testimonies while maintaining provenance.
Highlighted model performance and residual errors through systematic documentation and inspection.
Showed potential for reusable access to contested historical narratives.

Abstract

Abstract of the poster “Beyond Archives: Designing a Segmentation and ATR Workflow for Mid-Twentieth-Century Typescripts in a Contested-Memory Case Study”, presented at the “DARIAH Annual Event 2026: Digital Arts and Humanities With and For Society: Building Infrastructures of Engagement” in Rome, Italy, 26–29 May 2026.Poster ID: 125. The Beyond Archives project aims to develop a relational digital library for memory sources in contested historical settings, connecting institutional records and oral testimonies while keeping provenance, access conditions, and narrative contexts readable. The poster presents the written-stream workflow of the project, focusing on layout segmentation and Automatic Text Recognition (ATR) for mid-twentieth-century Italian typescripts from the Prefecture of Naples records. It shows how PAGE/ALTO outputs, TEI-oriented structuring, controlled correction, and paradata documentation can support reusable access while keeping model performance, residual errors, and remaining limitations inspectable.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper