Revisiting producing structured PDFs from LaTeX

Question

I am trying to produce a structured or 'tagged' PDF from pdftex (Tex Live 2013) that passes automated tests in Adobe Acrobat for tagging. These tests are the de facto indicator for a document accessibility, which are often required by universities, government agencies and others worldwide for published documents.

To test a document for accessibility / tagging, try this:

1. Open the PDF in Acrobat Reader
2. 'file -> 'document properties' -> 'Description'
3. under 'advanced', you'll see a field, 'tagged pdf'. Ideally it should say 'yes'. But it doesn't :(

Question: does anyone know of any way to create a PDF which passes Acrobat's 'Tagged PDF' test directly from LaTeX?

The ideal solution would be a package or two that can be called from the preamble of a latex document. I want something that is very low effort for the author, and can thus be very easily integrated with existing workflows.

Note: There has been discussion about this in the past (see tags and links on the right), but as of May 2014, there was no clear solution to this. Much of the existing discussion is from 2012 or earlier (see How can tagged PDFs be created that support Universal Accessibility and reflowing?), and so I'd like to see if we can kick-start this discussion.

Why this isn't a duplicate question: When I first posted this question in July 2013, it was flagged as a duplicate of a question from 2010. The answer references a presentation given at at TUG 2010. That presentation basically says "we're working on it", and is not an answer that allows me to implement a solution.

Because it is now 2014, I think it reasonable to expect that:

1. There have been new developments in this field through new packages or updates to core LaTeX
2. Attempting to comply with Section 508 will have required that people figure out solutions. This is a relatively recent requirement and so solutions may have changed since 2010
3. The software that is typically used to judge compliance (Adobe Acrobat) has changed several times since the question was not answered
4. Packages that address accessibility have been proposed but not come in to widespread use or even disappeared (e.g. the accessibility package, http://www.babs.gmxhome.de/da_ergeb.htm), and it is therefore reasonable to revisit this issue with fresh eyes.

Answer 1

This question has a lot of votes, but hasn't received an answer. So I am giving a ConTeXt solution. To create a tagged pdf, simply add:

\setuptagging[state=start]

at the top of your document. For example:

\setuptagging[state=start]

\starttext
\startsection
    [title={A section title}]
  \input ward
\stopsection
\stoptext

Then, pdfinfo test.pdf gives:

Title:          test
Creator:        ConTeXt - 2014.04.24 09:39
Producer:       LuaTeX-0.79.1
CreationDate:   Fri May 30 11:58:32 2014
ModDate:        Fri May 30 11:58:32 2014
Tagged:         yes
Form:           none
Pages:          1
Encrypted:      no
Page size:      595.276 x 841.89 pts (A4)
Page rot:       0
File size:      10992 bytes
Optimized:      no
PDF version:    1.6

Answer 2

Experience with the Accessibility Package

I downloaded the accessibility.sty style file from Babette Schalitz's website. This file is available under an LPPL license.

Using \usepackage[tagged]{accessibility} in my preamble allowed me you to generate a basic tagged PDF file that passes tests in Adobe Acrobat. However, the package didn't work if I needed to use roman numerals for the first page. This error seems to be because accessibility.sty uses displayed page numbers to build the document tag structure.

Updating the Accessibility Package

To fix the numbering problem I added the count1to package and replaced a few of the \pageref with \count1 in accessibility.sty. I've called this modified file accessibility-meta.sty for now, and posted it to GitHub. The package now seems to compile both articles and reports, and the output shows up as "tagged" in Acrobat.

Make sure that \usepackage[tagged]{accessibility} or \usepackage[tagged]{accessibility-meta} are pretty much the last thing in your preamble. In the event of a 'tex counter overflow, try commenting out various bits of your code. I've found that the tagging blows up quite regularly, especially with complex documents. Compile the document several times so that the page numbering and position of text or floats settles down, and the tags are properly generated.

Other Steps

I've also taken the steps that nbdb suggested:

1. Upgrading to TexLive-2015.
2. Adding \pdfinterwordspaceon to the preamble to fix the loss of inter word spacing in text (requires TexLive-2014 or later).
3. Adding the cmap package to fix mapping characters to unicode.

Testing and Feedback

I produced a test PDF using TexLive 2015, the cmap package, \ pdfinterwordspaceon, and the modified accessibility class. This seems to be "tagged". The PDF is available here.

Because I don't work with tagged documents on a regular basis it's highly likely I may not be doing a detailed enough test of the resulting document. I'd be keen to know if this PDF passes pre-flight testing in Acrobat, and if accessibility-meta.sty file works on other documents. I'm particularly interested in feedback from anyone who deals with tagged publications professionally.

Babett Schalitz deserves all of the credit for having produced the original accessibility.sty file, upon which accessibility-meta.sty is based. I would not have had a chance without her work.

Update August 2015: I am working towards getting accessibility-meta.sty on to CTAN. Please feel free to register issues on the github repository, or provide suggestions for improvements.

---

Update June 2020:

In 2019 I got in contact with Babett and got the files for the original accessibility package; she also allowed me to take over the maintenance of the package. I tidied up accessibility enough to get it to CTAN, but didn't update the functionality.

Releasing accessibility to CTAN has shown that unfortunately there are now quite some problems with it. When it was developed back in the early 2010s it worked a lot better, and it looks to be very sensitive to developments in other packages. It no longer compiles a basic MWE reliably.

Because of this I no longer think that accessibility is fit for purpose, and I will be contacting CTAN to look into getting it taken off CTAN (if possible).

However, I will leave the code at https://github.com/AndyClifton/accessibility. If anyone reading has coding skills and would like to contribute to the package, please leave an issue there.