Minutes_ MathML Full meeting_ 15 Jan_ 2026

Attendees:

  *   Neil Soiffer
  *   Louis Maher
  *   David Carlisle
  *   Paul Libbrecht
  *   Murray Sargent
  *   Deyan Ginev
  *   Moritz Schubotz
  *   Bert Bos
  *   Patrick Ion

Regrets

Action Items
2. Review PRs
ACTION: DC will finish the accent character and groupalign specifications.
ACTION: NS will start the horizontal review process.
ACTION: PL requests that issue 100 Simplification of the <semantics> element<https://github.com/w3c/mathml/issues/100> be included in next week's discussion.

Agenda

1. Announcements/Updates/Progress reports

DC reported on a new bug in the Firefox PDF reader supporting MathML tagging. it's the first browser-based, tagged PDF support, which I think is a big deal, because otherwise you've got to tell people to download the PDF and then open it in Acrobat or Foxy and stuff, and it's just not what people do these days.
The reader breaks if it tries to process a document with a table of contents. It is scheduled to hit the full release on the twenty-fourth of February; therefore, DC hopes that the reader can be repaired by that time.

DC: Once the Firefox reader ships, it is hoped that Chrome and Safari can be persuaded to improve their pdf capabilities.

DG: Anthropic announced yesterday that they have had their AI generate a new web browser, called FastRender, from scratch, Which is 3 million lines of code written by AI.
It has a MathML engine. DG plans to test this browser.

PL reported on an effort to get AI to produce a text-to-speech (TTS) engine. This person thought that this was not possible (https://mastodon.online/@fastfinge@interfree.ca/115837094890176762).

NS: My AI story is that I'm working on a method of converting MathML to Braille and vice versa using AI. There is not much Braille on the web. There are several Braille codes, and my AI is confusing the various codes with one another. I do not do backwards translation.

From Deyan Ginev to everyone: Maybe we should make an AI model that produces canonical MathML :-Easier

2. Review PRs

DC had a PR on malign which was in draft mode for a long time. He brought it out of draft mode.

DC also did another PR on the A tag. Because full wants to be a subset of core, and because core has the A tag element with several attributes, These elements got copied over to the MathML version of A in full. We said that A was just an alias of mrow.

NS cleaned up the mstyle text.

We had a discussion on aligning numbers on a page of equations. Intent has simplified the alignment process.

From Deyan Ginev to everyone: I may have first learned what a decimal point is during the time MathML 1.

DG: Simplifying is better. Question for MuS: can you do alignment with Unicode control characters?

MuS: I think it might be problematic, especially if we get into RDL math because you might want to use them as Arabic control characters.

ACTION: DC will finish the accent character and groupalign specifications.

ACTION: NS will start the horizontal review process.

MuS: Why did we decide to make the attributes case insensitive.

DC HTML has this.

3. Volunteers for proofreading the spec?

ACTION: PL requests that issue 100 Simplification of the <semantics> element<https://github.com/w3c/mathml/issues/100> be included in next week's discussion.

Zoom Math WG Intent Meeting Summary (2026-01-15)

Summary

The team discussed progress on various PRs and specifications, including updates to MathML documentation and attribute specifications. They explored challenges with AI-based text-to-speech engines and their usability for blind users, as well as issues with converting MathML to Braille and vice versa. The group agreed to simplify the MathML specifications, address open issues, and prepare for the next meeting by reviewing the specification and discussing potential agenda items.

Malign PR Progress Discussion

David and Louis discussed the progress on the malign PR, which David had promised to finalize over Christmas but didn't. He has since made improvements and undrafted an old PR to make it mergeable, though it still needs more work.

Firefox MathML PDF Reader Update

David reported progress on Firefox's PDF reader supporting MathML tagging, noting that while initial test cases passed, the feature broke when combined with a table of contents. The issue is scheduled to be fixed in Firefox version 148, which is set for beta release this week and full release on February 24th. Deyan shared an update about Anthropic's announcement of their AI-generated web browser called FastRender, which includes a 7,000-line Rust rendering MathML core, though he hasn't yet tested it.

AI TTS Challenges for Blind Users

The group discussed AI-based text-to-speech (TTS) engines, particularly focusing on their usability for blind users. Paul shared a Canadian developer's experience with Samuel Proux, who found AI TTS engines unusable due to content skipping and math errors, leading to a preference for classic TTS engines like Eloquence. Louis expressed his preference for Eloquence due to its reliability and stability over the past 20 years, despite its mechanical sound. Neil mentioned working on a paper about converting MathML to Braille and vice versa, noting challenges with AI's ability to handle low-resource languages like Braille, which lacks standardized encoding in many older documents.

AI Braille to MathML Challenges

The group discussed challenges with AI translation of braille to MathML, where Neil reported that current results are disappointing, with accuracy falling to 25-30% for longer braille sequences despite initial success with shorter examples. They explored potential solutions including creating an AI model for canonical MathML production and improving canonicalization processes, though Neil noted significant technical challenges and long processing times. The conversation concluded with Louis mentioning licensing issues for high-quality text-to-speech voices, particularly those from companies like Nuance, while Neil noted that Microsoft's voices are freely available.

HTML and MathML PR Updates

David and Neil discussed recent pull requests (PRs) related to HTML and MathML specifications. David explained that he had cleaned up and undrafted a PR for malign, and another PR for A tag had been approved and merged. They discussed how the A element in Full was added to match Core, including attributes like href, target, and download. David clarified that the type attribute, though present in the specification, is ignored by the system. He noted that the MathML version of the A element inherits the same attributes as the HTML version, making it easier for developers. David also mentioned that the schema needs to be updated to allow the A element, which is currently not digital.

Mstyle Documentation Attribute Clarifications

The team discussed changes to the Mstyle documentation, focusing on clarifying how attributes work and which elements they inherit. David suggested merging his PR with Neil's changes, which included reorganizing text into bullet lists and clarifying which attributes are supported in core versus extended MathML. They agreed to merge the PRs, with David noting that his own PR would have further changes after merging. The team also discussed the need to update the documentation regarding combining and non-combining characters, with David planning to make a pull request to the core specification to address this issue.

Feature Removal and Alignment Challenges

David and Neil discussed the removal of certain features in a draft, including groupalign, and the challenges this presented, particularly in maintaining compatibility with Word. David explained that the current draft removes these features but includes an example that relies on them, highlighting the need for a way to achieve decimal alignment. Neil inquired about how this would be done in LaTeX, to which David responded that LaTeX handles decimal alignment by adjusting the alignment of digits in a column, without splitting them into separate columns. Murray noted that the main reason for retaining these features was to maintain compatibility with LaTeX, not just Word.

MathML Alignment Discussion

The team discussed alignment issues in table cells and MathML, with David explaining that baseline alignment should be the default behavior unless specifically overridden. They debated the necessity of decimal alignment for polynomials and data tables, with Neil expressing a preference to drop it entirely while acknowledging existing software that generates it. David agreed to modify the PR that evening and suggested revisiting the wording and examples in the documentation.

MathML Specification Simplification Discussion

The team discussed simplifying MathML specifications, particularly around Maligngroup and alignment features. David committed to cleaning up the attribute specifications and removing unnecessary complexity, while Neil will begin horizontal review outreach after the changes are merged. The group agreed to address several open issues, including case-insensitivity in attribute values and potential Unicode control character alternatives. They noted that many PRs have been merged but not closed, and Patrick volunteered to review the specification for typos before a potential CR vote next week. Issue #100 regarding semantics element simplification was raised as a potential agenda item for the next meeting.