Explainer: Extended Descriptions for Publications and the Web

• Publishing Maintenance Group Accessibility Task Force

• Issue tracker: https://github.com/w3c/publ-a11y-extended-desc/issues

This explainer documents the need for and proposed semantics for extended descriptions: longer, structured content that explains images, diagrams, tables, formulas, and other information-rich material that cannot be adequately represented by brief alternative text alone but needs to be closely linked to the image it explains.

The document summarizes goals, non-goals, candidate approaches, examples, alternatives considered, and next steps to enable review by APA, TAG, implementers and the community.

Users encounter images and non-text content that convey complex information (e.g., technical diagrams, charts, mathematical notation, museum objects) where a short alt is insufficient and an extended, structured description is provided. Without clear affordances, users may not discover these descriptions or understand the linking between them and the primary content.

While there is a well-established standard for providing alternative text (alt attributes) for images, there is currently no equivalent standardized mechanism for extended descriptions. This inconsistency creates a gap in accessibility support: authors cannot reliably associate extended descriptions with their images, and assistive technologies cannot uniformly discover and surface them.

Authors and reading systems need a reliable, discoverable, and programmatic way to identify and surface extended descriptions in consistent ways without breaking reading flow or excluding non-AT users. Reading solution developers, in particular, require explicit semantics to offer a dedicated, optimized experience for accessing extended descriptions—whether through pop-ups, side panels, or other specialized UI patterns that preserve reading context.

Collection managers and accessibility experts need to be able to identify and collect extended descriptions with links to their context image.

At a platform level, this problem arises because extended image descriptions lack explicit, programmatically identifiable semantics. Web accessibility guidance requires that text alternatives and their relationships to non-text content be programmatically determinable, rather than inferred from presentation or author conventions.

HTML and ARIA follow this same architectural principle by providing explicit semantics for meaningful content, enabling consistent discovery and interaction by assistive technologies. While short text alternatives are programmatically associated with images, there is currently no equivalent semantic mechanism for extended descriptions. As a result, assistive technologies cannot reliably identify, announce, or expose extended descriptions consistently, even when authors provide them.

• Provide machine-discoverable semantics for extended descriptions.
• Support linking between primary content and descriptions, including across documents.
• Ensure reading systems can surface descriptions in a way that preserves pagination and context.
• Make descriptions usable by all readers, not just AT users.
• Enable reading systems to offer users the ability to hide or skip extended descriptions based on their preferences or reading context.
• Support text-to-speech and navigation features that can intelligently handle extended descriptions without disrupting reading flow.

• Mandate a single UI for presenting extended descriptions.
• Reintroduce problematic longdesc behaviors.

Publisher feedback from Brazil, Europe, and North America has informed this work, with proposed solutions reviewed and validated by some publishers. Publisher associations in Italy and France have been engaged. Community use cases and testing are documented in publishing and accessibility working groups (see References).

A proof-of-concept (POC) in both HTML and EPUB formats has been developed and refined over three years by the DAISY Transition to EPUB working group, demonstrating practical patterns and their effectiveness across reading systems. For more information on best practices, see the Extended Descriptions Best Practices.

Further user testing is recommended to validate discoverability and presentation patterns in paginated vs. continuous reading contexts.

Today, best practice relies on the use of aria-detailsto identify either a link to the extended description or the extended description itself. The aria-details attribute creates a programmatic relationship between an image and its extended description. The extended description can be implemented in two complementary ways depending on the publication format and authoring context:

Extended descriptions can be embedded directly before or after the image they explain. This approach works well for content that is primarily web-based or when authors prefer to keep all content in a single file. The aria-details attribute points directly to the description, with no need for intermediate links.

• The image should have a brief alt and an aria-details attribute pointing directly to the ID of the extended description container.
• Place the extended description in a container (that could be a details element or an aside) just after or before the image.
• The description container should have a unique ID matching the aria-details reference.
• If using <details>, the <summary> serves as the description heading.
• A backlink (role="doc-backlink") is not needed as the image is just before or after the extended description.

Example pattern:

<!-- Main content -->
<img id="img1" src="figure1.png" alt="Schematic of the device" aria-details="desc-img1">

<!-- Adjacent extended description -->
<details id="desc-img1">
    <summary>Extended description — Figure 1</summary>
    <p>...detailed structured description...</p>
</details>

Extended descriptions can be managed in a separate section rather than adjacent to the image. This approach avoids heavy additions to the original document structure and gives users the choice to consult the extra content without disrupting the reading flow.

• Place the extended description in a separate section (that could be in a different file).
• In the main content, after the image, add a link to the extended description. The link can be text or an icon (with accessible name).
• The image should have a brief alt and a aria-details attribute pointing to the link's ID.
• The link should have a unique ID.
• In case of an external file containing a series of descriptions, each description is in a section with a matching ID, a heading, a presentational copy of the image, the detailed description.
• A backlink (role="doc-backlink") must allow users to navigate back to the exact place they left in the main content.

Example pattern:

<!-- Main content -->
<img id="img1" src="figure1.png" alt="Schematic of the device" aria-details="extdesc-1">
<a id="extdesc-1" href="extended-descriptions.xhtml#desc-img1">Extended description</a>

<!-- Extended descriptions section -->
<section id="desc-img1">
    <h2>Extended description — Figure 1</h2>
    <img src="figure1.png" role="presentation" alt="">
    <p>...detailed structured description...</p>
    <a role="doc-backlink" href="chapter01.xhtml#extdesc-1">Back to image</a>
</section>

Both implementations use aria-details to create the semantic link between image and description. Authors should ensure the referenced content is visible to all users and test the pattern across common reading systems and screen readers because user agent and AT support can vary.

Tests revealed limitations related to identifying extended descriptions with certainty. It manifests in two critical areas:

• when a link points to an extended description, there is no standardized semantic marker to distinguish it from other types of links, preventing assistive technologies from announcing the link's purpose appropriately and limiting user agents' ability to provide specialized navigation or presentation features.
• extended description content itself lacks a specific semantic role to distinguish it from generic sections, making it difficult to automatically locate and process extended descriptions within a document or across a collection of documents.

Additionally, testing showed declarative workarounds being impractical when parsed using XPath, including identification limited to a single document, expensive preprocessing and DOM traversals, and a requirement for sophisticated parsing logic that makes implementations less scalable for many images or large collections — more error-prone and resource-intensive compared with simple, explicit markup approaches.

Footnotes provide a well-established, accessible pattern: a reference in the main flow that links to a separate, uniquely identified container with a backlink to return to the reading position.

DPUB ARIA’s doc-noteref / doc-footnote roles are a concrete example of how explicit, paired semantics enable assistive technologies to announce purpose, provide navigation, and allow tooling to extract related pairs programmatically. Thanks to this declarative semantics, reading solutions can enable specific UX features (which, for example, allow the user to keep their place in the text), but even if they do not, the system still works with links.

This precedent shows the value of: - unique IDs for references and targets, - bidirectional navigation (reference → note, note → backlink), - exposing semantics in the accessibility tree so user agents and AT can offer specialized affordances.

Extended descriptions share these link-and-return needs but differ in scope: they are typically longer, structured, and may include media or complex markup. That difference argues against reusing footnote roles directly.

The footnote model is a useful precedent for linking and navigation patterns, but extended descriptions benefit from distinct semantics (e.g., extendeddescriptionref / extendeddescription) that reflect their content and afford appropriate UX and tooling behavior. As they are links, they remain usable even with legacy reading systems.

A key benefit of explicit semantics is that reading systems can enable users to hide or skip extended descriptions based on their preferences. This is especially relevant when extended descriptions are adjacent to the image in the reading order, where they may otherwise disrupt the flow of the main text. Implementing hide/skip functionality relies on standardized markup that explicitly identifies image description content. Specific use cases include:

• Reading systems can allow users to show or hide extended descriptions, depending on their preferences or reading context.
• Read aloud and media overlays can skip extended descriptions during continuous listening, while still allowing users to access them when needed.

These capabilities ensure that extended descriptions enhance accessibility for users who need them while not imposing them on users who prefer alternative reading approaches or interaction patterns.

We propose two complementary ARIA roles to strengthen link semantics and make extended-description relationships explicit to assistive technologies:

• role="extendeddescriptionref" to mark the forward link (the anchor in the primary content that points to the extended description).
• role="extendeddescription" to mark the container that holds the extended description (which can be in the same document or in an external resource).

Example pattern:

<!-- Main content -->
<img id="img1" src="figure1.png" alt="Schematic of the device" aria-details="extdesc-1">

<!-- Adjacent extended description -->
<details id="extdesc-1" role="extendeddescription">
    <summary>Extended description — Figure 1</summary>
    <p>...detailed structured description...</p>
</details>

Example pattern:

<!-- Main content -->
<img id="img1" src="figure1.png" alt="Schematic of the device" aria-details="extdesc-1">
<a id="extdesc-1" role="extendeddescriptionref" href="extended-descriptions.xhtml#desc-img1">Extended description</a>

<!-- Extended descriptions section -->
<section id="desc-img1" role="extendeddescription">
    <h2>Extended description — Figure 1</h2>
    <img src="figure1.png" role="presentation" alt="">
    <p>...detailed structured description...</p>
    <a role="doc-backlink" href="chapter01.xhtml#extdesc-1">Back to image</a>
</section>

Using explicit markup reduces the computational overhead associated with reverse-checking aria-details to identify links to extended descriptions.

The combination of aria-details, role="extendeddescriptionref" and role="extendeddescription" provides bidirectional programmatic relationships even when extended descriptions reside in separate HTML files.

role="extendeddescription" would identify extended description containers across document boundaries.

Similar semantic identification challenges have been successfully addressed, demonstrating the value of specific semantic roles for different types of linked supplementary content. For example, DPUB ARIA roles provide the doc-footnote and doc-noteref roles to identify notes and their references, enabling assistive technologies and text-to-speech engines to announce them appropriately and user agents to implement specialized navigation features.

• Any ARIA role additions require coordination with the ARIA Working Group.

• ARIA exposes semantics to assistive technologies.
• Reading systems can implement strong affordances and a consistent user experience (for example, panels or a navigate-and-return flow).
• Tools can extract together images and extended description.

• <details>: native HTML, but problematic in EPUB due to pagination and inconsistent support.
• longdesc: rejected due to historical misuse, lack of adoption, and removal from HTML and EPUB specifications.
• rel="extendeddescription": does not allow the container of the extended description to be identified. Could provide additional context but is not considered by assistive technologies.
• epub:type: The use of epub:type is deprecated by the EPUB spec.
• RDFa/microdata: structured data approaches that add semantic richness but require parsing infrastructure and do not directly expose relationships to assistive technologies via the accessibility tree.
• Reusing doc-noteref/doc-footnote semantics: while structurally similar, extended descriptions differ in scope and purpose from footnotes; repurposing existing roles would create confusion for users and implementers.

• Accessibility: use aria-details, unique IDs, and visible links or icons. Backlinks help users return to their reading position.
• Internationalization: localize link text and icon alt text; mark descriptions with appropriate lang.
• Privacy/Security: treat description files as any other resource; respect EPUB packaging and web security policies.

• Browser developers, reading system developers and AT vendors should be engaged to validate UX and API exposure.
• The DAISY Consortium, Fondazione LIA and Benetech recommend the separate file technique for EPUB.

1. Socialize this explainer with APA, TAG, and the wider web platform community.
2. Discuss ARIA role proposals and aria-details usage with the ARIA Working Group.
3. Create samples HTML, EPUB and web publications demonstrating the recommended patterns.
4. Produce authoring guidance and lint rules to encourage correct usage and prevent misuse.
5. Coordinate with reading system and assistive technology vendors on implementation and UX affordances.

• Standardizing Extended Descriptions: User Stories, Testing Results, and Current Limitations
• Extended Descriptions Use Cases and Reading System Expectations
• Extended Descriptions Best Practices
• WCAG 2.2 — Non-text Content
• ARIA aria-details Attribute
• POC: Extended Description

• Adjacent examples (details and aside)
• In-file example
• External example (main content)
• External example (appendix)

This explainer has been written by Gautier Chomel summarizing previous works from Charles LaPierre and Gregorio Pellegrino, discussed and reviewed by Matthew Atkinson, Matt Garrish, George Kerscher, Steve Noble, Wendy Reid, James Yanchack, and others. All works conducted under Avneesh Singh coordination.