1. Self-Paced Learning
  2. LaTeX

LaTeX

This page will consist of instructions on how to make LaTeX accessible (a typesetting language used for research papers; particularly in STEM), along with best practices. The training will be split into the following sections:

  1. Overview
  2. LaTeX Best Practices and Authoring Rules
  3. LaTeX Accessibility Workflow
  4. Additional Resources

This training assumes you are already using LaTeX/TeX to author your materials, and serves as a guide for integrating accessibility standards into your current workflow.

If you are new to LaTeX and looking to learn more or get started, visit The LaTeX Project for an overview and basic examples.

Overview

The key components of this workflow are LaTeX and NVDA:

  • LaTeX – a typesetting language used for technical and scientific documentation; these are .tex files that typically get converted into PDFs
  • NVDA (NonVisual Desktop Access) – a free, open-source, portable screen reader for the Microsoft Windows operating system. This enables blind and vision-impaired individuals to use computers by translating the information on the screen into synthetic speech or braille.

What is the goal of this training?

The goal is to write LaTeX that can become semantic HTML with MathML in order to be accessible for NVDA screenreaders. This training will help with:

  • Showing researchers how to prepare LaTeX/TeX source for accessible conversion.
  • Converting STEM content into HTML5 with MathML (using Pandoc) so NVDA can read math continuously.
  • Demonstrating accessible text, equations, tables, charts, diagrams, and long descriptions.
  • Providing a workflow that can be posted on the website and used by faculty in classrooms.

Why does this workflow matter?

  • A visual equation is not enough for screen reader access.
  • The source file must preserve structure: headings, lists, tables, captions, links, and math.
  • HTML5 with MathML gives assistive technology semantic math to read and navigate.
  • The final test is not only visual appearance; it is whether NVDA reads the content correctly.
  1. Researcher writes accessible LaTeX
  2. Keep images, data, captions, and descriptions with the source files
  3. Convert to HTML5 + MathML using Pandoc
  4. Test with NVDA in Firefox or Chrome
  5. Publish HTML and provide source files for review
  6. Convert the HTML5 to PDF (optional)

LaTeX Best Practices and Authoring Rules

As mentioned, the goal is to author accessible LaTeX source code that converts seamlessly into semantic HTML with MathML, ensuring full accessibility for NVDA screen reader users. Note the following best practices and top 10 authoring rules when working with LaTeX to ensure accessibility. This section corresponds to step 1 of the Recommended Workflow (writing accessible LaTeX)

Top Best Practices for Researchers

  • Use sectioning commands such as \section and \subsection for structure.
  • Use real LaTeX math commands; do not insert equation screenshots.
  • Use captions for every figure and table.
  • Write long descriptions for charts, maps, diagrams, and complex images.
  • Use true data tables, not tables for visual layout.
  • Use descriptive links instead of “click here.”
  • Avoid custom macros unless the conversion workflow has been tested.

Below is an example of an accessible LaTeX snippet.

\section{Calculus Examples}

A derivative can be defined using a limit:
\begin{equation}
f'(x) = \lim_{h \to 0} \frac{f(x+h)-f(x)}{h}.
\end{equation}

Use standard LaTeX syntax whenever possible so the converter can produce clean MathML.

Authoring Rules

The converter understands predictable LaTeX better than heavily customized LaTeX. The preamble should support structure and links, not hide meaning inside style commands. Use common packages and avoid unusual setup when the goal is accessible conversion.

  • Use standard document classes and common packages whenever possible.
  • Keep layout packages simple; visual formatting should not carry meaning.
  • Use hyperref for real links, but make the visible link text descriptive.
  • Avoid custom macros until the conversion workflow has been tested.
A screenshot of LaTeX source code showing the document class and appropriate package structure.

The rule is simple: if it is math, write it as math. That is what allows Pandoc or LaTeXML to create MathML that NVDA can read and navigate. Equations should be source math, not screenshots or pasted images.

  • Use equation environments for displayed equations.
  • Use semantic math commands such as \lim, \frac, \sqrt, \int, and \sum.
  • Do not replace equations with images just because they look correct visually.
  • Keep explanatory sentences before or after equations so the reader has context.
  • NVDA should read inline math inside the sentence flow.
  • Displayed equations should be announced as math, not images.
  • Cases and inequalities should be read in a logical order.
  • Use ClearSpeak + Medium verbosity for a clearer demo voice when testing Math content.
A screenshot of LaTeX source code showing semantic math commands and properly defined equations and their explanations.

A screen reader user can jump by heading. That only works if the LaTeX source uses real sectioning commands instead of visual formatting. Headings become navigation points in the final HTML output.

  • Use \section and \subsection for document organization.
  • Do not create headings by manually bolding or enlarging plain text.
  • Write section titles that describe the topic, such as “Vectors and Matrices.”
  • Place the explanation before complex notation so users know what they are about to hear.
A screenshot of LaTeX source code showing a proper section with an appropriate description.

Matrix structure should be expressed with rows, columns, and standard delimiters. The converter can only preserve matrix structure when the source marks up the matrix structure. Visually aligned text is not enough.

  • Use environments such as bmatrix for matrix notation.
  • Use ampersands for columns and double backslashes for rows.
  • Do not draw matrices using spaces or images.
  • Define the variables around the equation so the notation has meaning.
A screenshot of LaTeX source code defining a matrix for a linear state-space model.

When the cases environment is used, NVDA can move through the piecewise function in a logical order. A screenshot of the same function would lose that structure. Piecewise logic needs a real cases environment, not a visual table or screenshot.

  • Use the cases environment for piecewise definitions.
  • Write each condition explicitly, such as u > u_max.
  • Keep inequalities as LaTeX symbols instead of plain visual approximations.
  • Explain what the function represents before showing the equation.
A screenshot of LaTeX source code with an example of piecewise definitions.

Greek letters and control notation must be explained in surrounding text. Accessibility is not only about whether NVDA can pronounce symbols. The paper should also explain what those symbols mean when they first appear.

  • Use standard commands for Greek letters and notation.
  • After an equation, define symbols such as P, Γ, and θ tilde.
  • Do not rely on the reader already knowing every symbol.
  • Use plain language to connect the equation to the research meaning.
A screenshot of LaTeX source code showing examples of adding Greek letters and notation using standard commands.

Lists should remain lists even when each item contains equations. If the LaTeX source uses a real list, the HTML output can preserve that list. That helps users understand process steps, algorithms, and procedures.

  • Use enumerate or itemize for real list structure.
  • Keep each step as a separate item so screen readers announce list position.
  • Use inline math inside list items when the equation belongs to the sentence.
  • Avoid using manual numbering typed into paragraphs.
A screenshot of the LaTeX source code to properly define a list.

A table is accessible when a reader can connect each cell to its header. The source should make the table simple and predictable before conversion. Tables need headers and should be used for data, not page layout.

  • Use a tabular environment for data with consistent rows and columns.
  • Include a caption that explains the table purpose.
  • Use a clear header row: Parameter, Symbol, Value, Meaning.
  • Avoid merged, nested, or layout-only tables when possible.
  • Do not rely on visual spacing to communicate relationships.
  • Test table navigation with T and Ctrl + Alt + arrow keys in NVDA.
A screenshot of the LaTeX source code used to create a true data table.

A chart should have a caption, a long description, and data when the data matters. For charts, the goal is not to describe every pixel. The goal is to provide the same academic meaning: what the chart shows, what changes, and what conclusion the reader should take away.

  • Use a figure caption for the visible chart.
  • Provide a long description that states the trend and main takeaway.
  • Add a data table when exact values are important.
  • Do not expect alt text alone to explain a complex chart.
  • Text equivalents for charts and diagrams – when to use what:
    • Use alt text for the short identification of the image.
    • Use a figure caption for visible and structural context.
    • Use a long description when the image contains important data or relationships.
    • For charts, also provide the key data or a linked data table when needed.

The examples below include a chart and a diagram, which should include a figure caption and a long description as the text equivalents, respectively.

Screenshot of a chart included in LaTeX that should have a caption.
Screenshot of a diagram included in LaTeX which should include a long description.
A screenshot of LaTeX source code defining an example stability chart with appropriate text equivalent.

A block diagram needs a written path through the system. When teaching diagrams, explain the process in order. A reader needs the same functional relationship that a sighted reader gets from the arrows.

  • Use the figure for sighted users and a text equivalent for screen reader users.
  • Describe the flow: input, summing junction, controller, plant, output, feedback.
  • Keep the long description close to the figure in the source file.
  • Avoid relying only on arrows or spatial position to communicate meaning.
A screenshot of LaTeX source code defining a block diagram of an example control-loop system with text equivalent.

LaTeX Accessibility Workflow

Once you have followed the LaTeX Best Practices and Authoring Rules, the next steps in the process are the accessibility workflow – ensuring your images, data, captions, and descriptions are located in the same place as the LaTeX source files, then proceeding with the Pandoc conversion to HTML5 + MathML in order to be able to test the content with NVDA screen reader. Once the HTML5 has been tested/verified, it can also be converted to a PDF file for distribution. This workflow corresponds to steps 2-6 of the Recommended Workflow.

Tools used in the Workflow

Local Tools (Recommended)

  • Pandoc: local converter used for the demo.
  • Command Prompt, PowerShell, or Terminal.
  • Firefox or Chrome for testing the HTML.
  • NVDA 2026.1.1 or later for math testing.

Web-Based Options

  • Try Pandoc: useful for simple math-only examples.
  • LaTeXML online upload: better for full projects with figures.
  • Online tools should not be used for confidential/unpublished research without approval.
  • For privacy, use local conversion.

Installing Pandoc Locally

1. Installing Pandoc on Windows

  • Start on pandoc.org, not an unofficial download site. Then navigate to “Installing” in the topnav.
  • For Windows beginners, choose the .msi installer because it handles setup automatically.
  • Avoid the .zip option for beginner instructions unless a manual install is required.
A screenshot of the official Pandoc download page.

2. Run the Windows Installer

  • Accept the license agreement, then click Install.
  • The default per-user install is usually enough and does not require administrator privileges.
  • After the installer finishes, close and reopen Command Prompt before testing Pandoc.

Accept license

A screenshot of the license agreement that the user gets prompted to accept when installing Pandoc.

Install progress

A screenshot of the download screen while waiting for the Pandoc Windows installation.

Finish setup

A screenshot of the completed installation confirmation.

3. Verify the installation

Before proceeding with any LaTeX conversion, confirm that the system recognizes the Pandoc command

  • Open a new Command Prompt window after installation.
  • Run: pandoc --version
  • A version number confirms Pandoc is installed and available on the system path.
  • If the command is not found, reopen Command Prompt or restart the computer.
A screenshot of the successful Pandoc installation via Windows terminal.

1. Installing Pandoc on macOS

Installing Pandoc using Homebrew (recommended)

Homebrew is a package manager for macOS that can help you easily and quickly download software and packages for your Mac. Read more about Homebrew and how to install it.

Once Homebrew is installed, to install Pandoc using Homebrew, simply open up a new terminal instance and run the following command:

brew install pandoc

Installing Pandoc using the official download page

Visit Pandoc’s download page for macOS and follow the steps to use the latest package installer on Pandoc’s download page.

2. Verifying the Installation

Before proceeding with any LaTeX conversion, confirm that the system recognizes the Pandoc command

  • Open a new Terminal window after installation.
  • Run: pandoc --version
  • A version number confirms Pandoc is installed and available on the system path.
  • If the command is not found, reopen Terminal or restart the computer.
A screenshot of the macOS Terminal prompt verifying the Pandoc installation and outputting version 2.12.

Prepare the Project Folder

The Pandoc converter needs the .tex file and every resource referenced by the LaTeX source. The following section applies to LaTeX project folders in both Windows and macOS.

  • Keep the .tex file and the figures folder together in one project folder.
  • Use relative image paths in LaTeX, such as figures/stability_plot.png.
  • Do not rename image files after writing the LaTeX source.

Note that if the path in LaTeX does not match the folder, the image will not appear in the converted HTML.

Example:

A screenshot of the required file setup for the resources referenced by the LaTeX source that is getting converted.

The LaTeX command \includegraphics{figures/stability_plot.png} requires a real figures folder with that exact file name, as seen on the left.

Convert LaTeX to HTML5 With MathML

Once the project structure is prepared per the above, you can start using Pandoc locally to convert your LaTeX resources into accessible HTML5 for use with NVDA.

Note: before running any commands, ensure that you cd into the correct directory via Command Prompt on Terminal where your .tex LaTeX source files and supplemental resources in the project folder are located. (Example: cd /Users/Myriam/Library/CloudStorage/OneDrive-UniversityofConnecticut/Documents/latex/)

Use the following local conversion command in Command Prompt, replacing accessible_math_paper_demo.tex with your .tex original LaTeX source document, and accessible_math_paper_demo_mathml_standalone.html with the desired name for your new output HTML5 file:

pandoc accessible_math_paper_demo.tex -s --toc --mathml --embed-resources -o accessible_math_paper_demo_mathml_standalone.html
  • -s creates a full standalone HTML document.
  • --toc creates a linked document index/table of contents.
  • --mathml converts LaTeX equations into MathML.
  • --embed-resources embeds images into the HTML file.
  • -o names the output file.
A screenshot of the Windows command prompt for the conversion using Pandoc.

In Terminal, first run pandoc --version to determine which command you will use for the conversion.

Screenshot of a Terminal prompt on Mac checking the Pandoc version.

If you have a Pandoc version older than 2.19

Use the following local conversion command in Terminal, replacing accessible_math_paper_demo.tex with your .tex original LaTeX source document, and accessible_math_paper_demo_mathml_standalone.html with the desired name for your new output HTML5 file:

pandoc accessible_math_paper_demo.tex -s --toc --mathml --self-contained -o accessible_math_paper_demo_mathml_standalone.html

If you have Pandoc version 2.19+

Use the following local conversion command in Terminal, replacing accessible_math_paper_demo.tex with your .tex original LaTeX source document, and accessible_math_paper_demo_mathml_standalone.html with the desired name for your new output HTML5 file:

pandoc accessible_math_paper_demo.tex -s --toc --mathml --embed-resources -o accessible_math_paper_demo_mathml_standalone.html
  • -s creates a full standalone HTML document.
  • --toc creates a linked document index/table of contents.
  • --mathml converts LaTeX equations into MathML.
  • --embed-resources embeds images into the HTML file for Pandoc versions 2.19+.
  • --self-contained embeds images into the HTML file for Pandoc versions older than 2.19.
  • -o names the output file.
A screenshot of the command on a Mac terminal to cd into the correct directory, check the Pandoc version, then use the respective Pandoc command to convert .tex files into an HTML5 output.

You can also use Pandoc on the web to convert, although it is not recommended because it is not able to integrate images or figures. Additionally, online tools should not be used for confidential/unpublished research without approval. For privacy purposes it is better to use the local versions mentioned above (for Windows and macOS). However, Pandoc online can be useful for simple, math-only examples.

Use the following link to access Pandoc on the web.

  • Select from: LaTeX and to: HTML5.
  • Check Standalone and TOC.
  • Choose MathML for math output.
  • Paste LaTeX or upload a simple .tex file.
  • If figures do not load, the uploaded file names and image paths likely do not match.
A screenshot of the Pandoc on the web interface with the appropriate options selected.

Note on MathML vs. MathJax for the output:

MathML HTML (Recommended)

  • Recommended first output for NVDA testing.
  • Math is embedded semantically in the HTML.
  • Works well with NVDA 2026.1.1 in our test.
  • Use –mathml in Pandoc.
  • Select MathML for math output on Pandoc on the web

MathJax HTML

  • Useful comparison output for visual rendering.
  • Often depends on JavaScript and internet access unless configured locally.
  • Screen reader behavior can vary by browser and settings.
  • Use –mathjax in Pandoc.
  • Select MathJax for math output on Pandoc on the web

Setting up NVDA

Once the LaTeX has been converted to HTML5, the NVDA screenreader can be used for testing its accessibility.

Note: NVDA is only available on Windows and incompatible with macOS at this time. For more information on testing NVDA on a Mac and possible workarounds, see here.

Download NVDA for Windows

Once installed, note the following settings and instructions for NVDA:

  • Use NVDA 2026.1.1 or later.
  • Math speech style: ClearSpeak.
  • Speech verbosity: Medium.
  • Use Firefox first, then Chrome if needed.
  • Keep Browse Mode active on the HTML page.
Screenshot of the NVDA version and settings for Windows
Screenshot of the NVDA recommended settings for the NVDA screenreader including speech style and speech verbosity.

Document structure

  • NVDA + Down Arrow: continuous reading
  • H / Shift + H: next or previous heading
  • Tab / Shift + Tab: index links
  • NVDA + F7: list headings and links
  • G: graphics/images

Math and tables

  • T / Shift + T: next or previous table
  • Ctrl + Alt + Arrow keys: table cells
  • NVDA + Alt + M: math navigation
  • Arrow keys: move inside math
  • Escape: leave math navigation

  • Document Index links jump to the correct sections.
  • Heading navigation follows the intended structure.
  • Continuous reading moves from text into math and back to text.
  • Inline and display equations are read meaningfully.
  • Matrices, cases, Greek symbols, and inequalities are understandable.
  • Tables can be navigated by rows and columns.
  • Images, captions, and long descriptions are read in order.
  • The output works in Firefox and is checked in Chrome if possible.

Converting HTML5 to PDF

Once your LaTeX has been converted to HTML5 and successfully tested with NVDA, you can then convert the HTML5 files to PDF for distribution. The following instructions can be used for Windows and macOS:

Troubleshooting the LaTeX Accessibility Workflow

  • Confirm the file path matches /includegraphics.
  • Run the command from the project folder.
  • Use --embed-resources for a single-file HTML output.
  • For online conversion, use a ZIP workflow like LaTeXML.

  • Confirm the HTML output uses MathML.
  • Use NVDA 2026.1.1 or later.
  • Use Firefox first.
  • Check Math settings: ClearSpeak, Medium verbosity.
  • Avoid equation screenshots or unsupported custom macros.

Additional Resources