The “Larger Example” on the WTPDF from LaTeX page shows a small typical LaTeX article set in two columns with tables and figures.
Here we show the effect of tagging on generated speech. All examples use the test release of NVDA 2025 (enabling interfacing to MathML readings), along with public versions of MathCat, Foxit and Acrobat Reader on Windows 11.
To present something of the experience to users not familiar with screen readers, we have provided video clips of readings of the full document and of specific sections. Although a video format is used they are all recordings of the screen reader, with a fixed video frame showing the part the document in each case.
The first overview of a document is often presented by a table of contents. However a screen reader may have difficulty inferring a usable reading. With an untagged document it relies on the screen reader inferring the structure, to pass to the underlying Accessibility tools.
On this document foxit produces the following reading (Acrobat is similar).
Using the tagged PDF, the structure is made explicit to the screen reader and typesetting atrtifacts such as leader dots are marked as artifacts and not read, producing a far more understandable reading.
Mathematics is traditionally hard to automatically present.
In the untagged PDF the screen reader is just passed the raw character data. The superscript in the first equation is not read, so losing the meaning, and the matrix equation is completely scrambled with the entries being read in the wrong order.
At the present time different PDF readers support different ways of tagging mathematics, here we present a version with math tagged with Associated files containing MathML, which is understood by foxit, and a version with the same MathML used as PDF Structure Element tagging, which is understood by Acrobat. Identical readings are generated as in either case the resulting MathML is passed via NVDA to MathCat to generate the reading.
We expect that this is a temporary situation and that PDF readers will soon support both mechanisms to supply MathML to the accessibility tools. However for now, we need to build two separate PDFs, one using Associated Files, and one with Structure Elements, to support current PDF viewers.
Associated Files and foxit
Structure Elements and Acrobat
Available in three forms
t1, Classic LaTeX, typeset with pdftex and no tagging enabled.
TeX
PDF
t1-af Lualatex with Unicode Math and tagging enabled. Auto-generated MathML included as Associated files.
TeX
PDF
t1-se Lualatex with Unicode Math and tagging enabled. Auto-generated MathML included as PDF Structure Elements.
TeX
PDF