Scaling Accessible Mathematics on arXiv: HTML Conversion and MathML 4
This paper reports on the 2025–2026 progress of arXiv's experimental HTML Papers initiative, highlighting community-driven fidelity improvements, corpus-scale conversion efforts, the introduction of MathML 4 for accessibility, and a Rust-based port of LaTeXML to enhance performance and reduce costs.
Original paper dedicated to the public domain under CC0 1.0 (http://creativecommons.org/publicdomain/zero/1.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
The Big Picture: Turning "Frozen" Math into "Living" Text
Imagine the world's largest library of scientific papers (arXiv) is currently filled with books printed in a very specific, rigid ink. These books look beautiful and precise (PDFs), but if you try to read them with a screen reader (for the blind), zoom in on a tiny diagram, or ask a computer to summarize the math, the ink doesn't move. It's "frozen."
The authors of this paper are building a new system to turn that frozen ink into "living text" (HTML). This living text can flow to fit any screen size, be read aloud by computers, and be understood by search engines. Their goal is to make sure that when a scientist writes a math paper, it works for everyone, not just people with perfect eyesight and a specific browser.
The Four Main Achievements (The "Highlights")
1. Fixing the Leaks (Community Help)
Think of their new system as a new boat they launched in 2023. Like any new boat, it had some leaks. Over the last year, they listened to the passengers (readers and authors) who reported where the water was coming in. They fixed about half of the 6,000 reported problems. They are now working on making the boat watertight enough that 90% of all papers can sail through without sinking.
2. The "Intent" Label (Teaching Computers to Speak)
Math is hard for computers to read aloud because symbols can mean different things. For example, a small "x" could be a variable, a multiplication sign, or a letter.
- The Old Way: The computer guesses, often getting it wrong and sounding robotic or confusing.
- The New Way (MathML 4): The authors are adding "intent labels" to the math. It's like putting a sticky note on a word that says, "Say this exactly as written" or "This is a specific scientific term." This helps screen readers speak the math clearly, even if the symbols are weird or new.
3. The Rust Engine (Speeding Up the Factory)
The system that converts the papers is currently built with an old programming language called Perl. It's like a factory running on a slow, aging engine.
- The Problem: The Perl engine is slow, expensive to run, and fewer people know how to fix it anymore.
- The Solution: They are rebuilding the engine using a modern language called Rust.
- The AI Twist: Usually, rewriting a massive engine by hand takes years. However, they used a very advanced AI (called "Agentic AI") to help write the code. The AI acted like a super-fast apprentice, writing about 100,000 lines of code in just three weeks—work that would have taken two humans two years to do.
- The Result: The new Rust engine is 10 to 30 times faster. This means authors can see their papers converted to the new format almost instantly when they submit them.
4. The "Live" Pipeline
They aren't just fixing the new papers; they are building a pipeline to eventually convert the 3 million old papers in the archive. Right now, the system is experimental, but they are steadily improving it so that eventually, every paper in the library will be accessible.
How They Know It's Working
They use two ways to check their progress, like a mechanic checking a car:
- The "Squeak" Report: They look at complaints from users. If a user says, "This diagram looks broken," they fix it immediately.
- **The "Stress Test": They run the system over the entire history of papers to see which math formulas or packages cause the system to crash. This helps them find hidden problems that no single user might have reported.
The Future
The authors believe that in the future, writing a math paper that is accessible to everyone should be the default, not a special extra step. They are also in the middle of a big change for their organization (becoming an independent non-profit), which they hope will give them the stability to keep building these tools for the long haul.
In short: They are taking the world's biggest math library, upgrading the "reading glasses" so everyone can see the math clearly, and using AI to build a faster, cheaper engine to make it all happen.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.