#1205: WG Revision: MathML 4

Visit on Github

Opened Mar 19, 2026

Specification

https://www.w3.org/TR/mathml4

Explainer

https://w3c.github.io/mathml-docs/intent-explainer/

Links

  • The WG's request for this TAG review: https://lists.w3.org/Archives/Public/www-math/2025Dec/0007.html
  • TAG review of the previous version of this specification, if any: MathML 3 (CR in 2009) apparently predates TAG review, I couldn't find anything. MathML Core's review is at: https://github.com/w3ctag/design-reviews/issues/438.
  • A description of what has changed since our previous review: There are two major changes:
    1. Alignment with MathML Core, which was reviewed before it went into CR.
    2. The addition of "author intent". These are attributes that can be used by authors to disambiguate the speech for a notation. The goal is improved accessibility. This is the significant change that should be reviewed.
<!-- Make a copy of the below items for each distinct feature that has changed since our previous review, and fill in links to describe that feature. For small changes, these may all be included in the issue that motivated the change, but please double-check that all the answers are actually there. -->

Feature 1:

  • Previous early design review, if any: None. There were discussions around 2022 with the ARIA WG that led to the conclusions in the explainer about ARIA not being an appropriate solution.
  • An introduction to the feature, aimed at unfamiliar audiences: see the early parts of the explainer.
  • A description of the problems that end-users were facing before this proposal: https://w3c.github.io/mathml-docs/intent-explainer/#semantic-reading-considerations
  • Alternatives considered: https://w3c.github.io/mathml-docs/intent-explainer/#leveraging-existing-technology
  • Examples of how to use the proposal to solve the end-users' problems: a few simple examples are shown in the explainer. A much larger set of examples can be found here. Note: the speech in the larger set of examples comes from MathCAT, which is used by the Windows-based screen readers JAWS and NVDA among others. MathCAT does far more inference of speech than other accessibility solutions, so the difference between the default reading and the intent reading is much smaller than is the case in other screen readers. For example, $P^n_k$ is not likely to be read as "k permutations of n" by most other MathML-to-speech solutions. They are more likely to read it like "p subscript k superscript n end superscript" or (incorrectly) as "p sub k to the nth power".
  • What do the end-users experience with this proposal: see above answer with links to examples of speech.
  • User research you did to validate the problem and/or design, if any? As mentioned in the explainer, there are no studies showing semantic speech is better than literal speech. But it is typically shorter (hence, it uses less working memory) and it corresponds to what is spoken in the classroom and elsewhere. So it is generally believed to be preferable. Validation of the design was done by examining many examples along with making sure it could both be generated by authoring tools and consumed by assistive technology. In particular, the addition of intent properties solved a long-standing problem of how to speak some tabular layouts well. These are often represented as tables to align parts of an equation such as an equals sign and otherwise get spoken with "row 1, col 1: ..., col2 ..." when in fact they are not tabular.
  • Web Platform Tests: MathML Core has Web Platforms Tests. There are no platform independent tests for intent because the spec intentionally does not require a specific way to speak the math. Different speech is appropriate for different audiences (see the explainer).

The specification

Where and by whom is the work is being done?

  • GitHub repo: https://github.com/w3c/mathml/issues
  • Primary contacts:
    • Neil Soiffer (@nsoiffer), Talking Cat Software, Math WG Co-chair
    • Brian Kardell (@bkardell), Igalia, Math WG Co-chair
  • W3C Math WG
  • This work is being funded by: None. Volunteer-based with each authoring and AT vendor self-funding their implementation
  • Primary standards group developing this feature: Math WG
  • Incubation and standards groups that have discussed the design:
    • sorry, I can't find the discussion links with ARIA and internationalization groups

Feedback so far

  • Active horizontal reviews:

  • Privacy: https://github.com/w3cping/privacy-request/issues/197

  • Security: https://github.com/w3c/security-request/issues/121

  • Accessibility: https://github.com/w3c/a11y-request/issues/151

  • Internationalization: https://github.com/w3c/i18n-request/issues/299

  • Multi-stakeholder feedback:

    • Browsers only pass the intent and arg attributes through, they do not process them and so no comments.
    • The following all are happy with and contributed in the development of intent:
      • MathCAT is used by many Windows-based assistive technology (JAWS, NVDA, Dolphin EasyReader, Kurzweil 3000, ...) and it has implemented most of the intent specification
      • The LaTeX project, which maintains TeX/LaTeX has implemented some intent generation and have added a macro to generate intent. They plan to incorporate that macro into many of the LaTeX packages
      • The Wikimedia foundation has incorporated intent into one of its MathML editors

  • Adobe Acrobat/Reader, Foxit Editor/Reader, and Firefox PDF's viewer all support MathML in PDF and pass along the intent attributes.

  • Major unresolved issues with or opposition to this specification: none, but the Math WG continues to refine the core concept list and properties based on usage and feedback. In submitting this review, I became aware of the registries and the Math WG will look into using that for intent tables.

  • Status/issue trackers for implementations: none yet

You should also know that...

As a reminder, the only new addition besides aligning with MathML core, is intent, which is a small part of the spec.

The Math WG has started the re-chartering process. Our charter ends in April. It would be nice, but not essential, to start the new WG (assuming the group is re-chartered) with the reviews in hand so we can move to CR and move that stage along.

Discussions

Log in to see TAG-private discussions.

Discussed Apr 27, 2026 (See Github)

Jeffrey: This intent attribute overlaps with the semantic MathML elements. The explainer says that the semantic elements haven't been adopted in the past 10 years. But they need to say why they think that, rather than proposing a new thing, which we don't know the changes of being adopted.

Brian: Effectively, there was a tool with "presentational MathML" to do what this is trying to do: more or less, to read it correctly. It's unclear how to read some math formulas. Everyone uses the presentational one. The semantic one is alittle deeper in the weeds. It's a little more "semantic web-y" as opposed to light annotation to clarify. The tools that have been used all sort of do heuristic matching. This is used by JAWS and other plugins. The idea with the "intent" is to ust be able to add some clarification with some light annotation. It's more like a presentational hint than a required thing. You can make it better. It is currently being used; there is a tool, MathCad [sp?], from Microsoft. I think it definitely stands a better chance of success. That said, I'm not a part of it. We've said "these things are real and in the browser, and these are the speculative things." They have focused almost exclusively on this "intent" attribute, and I get it. The idea, though, is that if it takes off, then maybe we can get it into MathML core in the next version. We have even reserved it as an attribute that could be introduced later.

Luke: My question was going to be which of the designs seems more in-keeping with MathML 4 and the wider web platform, rather than the XSLT version of the web. If intent is what is being used in real MathML in the wild, and may be included in MathML Core, maybe that answers the question of what's different between the semantics stuff specified before and what's proposed here. Haven't looked into the detiails but concern about the HTML design - maybe an attribute is OK.

Brian: it's a hint, like ARIA, though it has its own format.

Luke: If you were designing this from scratch, would you make it accessibility-specific i.e. part of ARIA, or does it have wider uses?

Brian: Its primary use is accessibility. It's very well thought-out in a lot of ways. But semantics are semantics. Being able to understand them makes it a better target for AI and other applications. If you can convey semantics, then others can use them.

Jeffrey: It looks like this is not useful for evaluating the math, but this is for presenting it to people. It does seem like they should at least talk to the ARIA WG about how this will integreate with the rest of the accessibility annotation systems. I think you said that MathCat has implemented this; have other screen readers implemented it?

Brian: I know he is working with people from those other groups. He has attended some ARIA meetings. He was at CSUN recently. He's working with a lot of groups. I think they're generealaly on-board. We tried to bring them together when the proposal for HTML was a thing. (It does some wild emedding of SSML in HTML. It's interesting, but it's complicated to serialize.) That was mostly the pronounciation task force. Matthew was involved in that. We tried to bring that to the same group. I encouraged Neil to attend, and we even tried to link them with James Craig who had similar ideas--maybe we do an IPA attribute or something liek that. There are a lot of related questions and problems. They're all a little bit different...

Matthew: https://w3c.github.io/mathml-docs/notes-on-mathml/#notes-on-mathml-accessibility https://w3c.github.io/mathml-aam/ I was going to ask about how widely-used and widely-supported "intent" was. I have a sense it hasn't been implemented yet. I found somethings outside the scope of this document. They propose to have a "note track" document. It will include a number of things, including a section on accessibility (linked above). It will evolve to include best-practices. There is an AAM document being proposed; Neil and James are involved in it. I don't know how well-supported this is, but they're definitely talking with the right groups.

Brian: the AAMs are more for MathML Core. It's not in MathML core, nbut it could be in the future. I guess the AAMs would work regardless. Probably? The AAMs are generally for binding in HTML.

Jeffrey: I like the syntax they've come up with. Sounds like they've been and are talking to the right people. One tweak with how they're arranging these documents: they have these side documents that talk about the terms they're defining. These should probably be W3C Registries, rather than general side documents.

Brian: So we can say 'continue talking to the right people'

Lola: Leaving it with Jeffrey.