What is the purpose of the "role" attribute in HTML?

As I understand it, roles were initially defined by XHTML but were deprecated. However, they are now defined by HTML 5, see here: https://www.w3.org/WAI/PF/aria/roles#abstract_roles_header

The purpose of the role attribute is to identify to parsing software the exact function of an element (and its children) as part of a web application. This is mostly as an accessibility thing for screen readers, but I can also see it as being useful for embedded browsers and screen scrapers. In order to be useful to the unusual HTML client, the attribute needs to be set to one of the roles from the spec I linked. If you make up your own, this 'future' functionality can't work - a comment would be better.

Practicalities here: http://www.accessibleculture.org/articles/2011/04/html5-aria-2011/

Is this role attribute necessary?

Answer: Yes.

• The role attribute is necessary to support Accessible Rich Internet Applications (WAI-ARIA) to define roles in XML-based languages, when the languages do not define their own role attribute.
• Although this is the reason the role attribute is published by the Protocols and Formats Working Group, the attribute has more general use cases as well.

It provides you:

• Accessibility
• Device adaptation
• Server-side processing
• Complex data description,...etc.

Most of the roles you see were defined as part of ARIA 1.0, and then later incorporated into HTML via supporting specs like HTML-AAM. Some of the new HTML5 elements (dialog, main, etc.) are even based on the original ARIA roles.

http://www.w3.org/TR/wai-aria/

While the First Rule of Aria states:

If you can use a native HTML element [HTML51] or attribute with the semantics and behavior you require already built in, instead of re-purposing an element and adding an ARIA role, state or property to make it accessible, then do so.

there are a few primary reasons to use roles in addition to your native semantic element.

Reason #1. Overriding the role where no host language element is appropriate or, for various reasons, a less semantically appropriate element was used.

In this example, a link was used, even though the resulting functionality is more button-like than a navigation link.

<a href="#" role="button" aria-label="Delete item 1">Delete</a>
<!-- Note: href="#" is just a shorthand here, not a recommended technique. Use progressive enhancement when possible. -->

Screen readers users will hear this as a button (as opposed to a link), and you can use a CSS attribute selector to avoid class-itis and div-itis.

[role="button"] {
  /* style these as buttons w/o relying on a .button class */
}

[Update 7 years later: removed the * selector to make some commenters happy, since the old browser quirk that required universal selector on attribute selectors is unnecessary in 2020.]

Reason #2. Backing up a native element's role, to support browsers that implemented the ARIA role but haven't yet implemented the native element's role.

For example, the "main" role has been supported in browsers for many years, but it's a relatively recent addition to HTML5, so many browsers don't yet support the semantic for <main>.

<main role="main">…</main>

This is technically redundant, but helps some users and doesn't harm any. In a few years, this technique will likely become unnecessary for main.

Reason #3. Update 7 years later (2020): As at least one commenter pointed out, this is now very useful for custom elements, and some spec work is underway to define the default accessibility role of a web component. Even if/once that API is standardized, there may be need to override the default role of a component.

Note/Reply

You also wrote:

I see some people make up their own. Is that allowed or a correct use of the role attribute?

That's an allowed use of the attribute unless a real role is not included. Browsers will apply the first recognized role in the token list.

<span role="foo link note bar">...</a>

Out of the list, only link and note are valid roles, and so the link role will be applied in the platform accessibility API because it comes first. If you use custom roles, make sure they don't conflict with any defined role in ARIA or the host language you're using (HTML, SVG, MathML, etc.)