Radiance: studying solutions for image description in scholarly backlist collections

In 2023, the abelab.eu project identified the accessibility needs of existing backlists, recognised the importance of generative AI, and emphasised that it must be integrated with human oversight and professional evaluation. Since then, various stakeholders—publishers, service providers, and distributors—have begun implementing these technologies and providing user-friendly interfaces.

However, significant challenges persist: understanding AI’s potential, documenting its applications, and developing relevance markers to guide human operators. Our mission is to equip the entire ecosystem with the necessary tools, guides, and connectors while supporting the professionalisation of practices.

We are taking a step forward with Radiance, a collaborative project that will enable study the use of AI to improve the accessibility of images in scholarly and academic digital publications.

The Challenge

Meaningful alternative text descriptions are crucial for making visual content accessible to readers who are blind or visually impaired. Nevertheless, creating effective alt text for the thousands of images in academic and scholarly works—from diagrams and charts to historical photographs and illustrations—is a massive, often overwhelming task.

FeniXX operates a unique backlist collection of nearly 70,000 titles in modern format, making it an ideal setting for studying how to address the needs of people with disabilities with minimal financial resources. The books that will be selected as part of this project will constitute a representative sample of scholarly books published in the last century.

Isako brings a wealth of expertise in content structuring and accessibility, particularly in academic and scholarly digital content. They will focus on ensuring that the solutions developed are not only technically sound but also scalable to meet the needs of large collection owners.

EDRLab’s Role

As a partner in Radiance, EDRLab brings its expertise in reading system technology and digital publishing infrastructure. We are contributing our Readium CLI analysis tool, which has been improved in the Soliman project. This tool extracts metadata and images from publications, along with their sizes and unique identifiers. We will enhance it by adding contextual information, such as captions, nearby paragraphs, and existing metadata. This context is crucial for generating accurate and meaningful alternative text descriptions.

EDRLab is also involved in defining the technical specifications and APIs that will enable this accessibility workflow to be integrated into existing publishing pipelines. Additionally, we will develop a component for a web reading solution that allows users to consult, modify, evaluate, and validate the generated descriptions.

Our goal is to create open-source tools and standards that the entire publishing industry can adopt, fostering scalable and sustainable accessibility remediation.

Why This Matters to EDRLab

This project supports our 2024 ABELab impact recommendations and paves the way for the trustworthy use of AI-generated content across large collections. The insights gained will help the entire profession implement generative solutions based on clear, publicly available guidelines and reference implementations.

Through our collaboration with Isako and FeniXX, we aim to advance these important objectives with a real-life industry use case.

Qualebook

Digital publishing has developed appropriate industrialization and standardization processes, such as the EPUB format, but is not immune to external injunctions linked to legal provisions and societal issues.

To remain profitable while meeting the expectations of audiences and funders, publishing houses need to acquire a wide range of skills. Digital production teams cannot confine themselves to typography, performance, interoperability, accessibility, eco-design, security or personal data management. It is imperative to coordinate these diverse areas of expertise.

All digital publishing professions are involved, and the challenges are in terms of decision-making, risk management, training, methods, resources and governance. To integrate and assimilate diverse outside expertise, digital publishing needs to equip itself with quality assurance, management and control tools for design, production and distribution, as well as for training and skills certification.

Qualebook website

News about Qualebook

Soliman:
Optimising Unavailable Books
and their Metadata for Digital Accessibility

Since 2014, FeniXX has digitised nearly 98,000 20th-century “Unavailable Books” (ReLIRE register), mainly in EPUB3 format, with 71% of its catalogue. In 2019, FeniXX began integrating WCAG standards into its production to achieve native accessibility. Despite potential exemption due to its size, FeniXX aims to implement accessibility adaptations that aren’t disproportionately burdensome for its heritage catalogue. FeniXX consulted EDRLab for their digital accessibility expertise.

EDRLab will develop an open-source tool to analyse digital book corpora, extract and inject accessibility metadata, and generate metadata files. Both FeniXX and EDRLab will make the tool and documentation openly available to benefit the digital book sector.