PDF language tagging (a short text)

[jbezos]

To be updated. See What’s new in babel 26.14.

Version 25.12 of the manual included a tentative example for a very basic and minimal tagging of text with \foreignlanguage (only text, without a par break). It was incomplete and didn’t work. Here a more complete one, but note proper tagging in complex situations is far from trivial.

Be aware the behavior related to tagging can and very likely will change.

\DocumentMetadata{lang=en, tagging=on, pdfstandard=ua-2}

\title{Test}

\documentclass{article}

\usepackage[welsh]{babel}

\ExplSyntaxOn
\AddBabelHook[welsh]{meta}{foreign}{
  \BabelWrapText{
    \tag_mc_end_push: 
    \tagstructbegin{tag=Span, lang=cy-GB}
      \tagmcbegin {} ##1 \tagmcend
    \tagstructend
    \tag_mc_begin_pop:n {}}}
\ExplSyntaxOff

\begin{document}

before
\foreignlanguage{welsh}{\localename\ \emph{emphasis}}
after

\end{document}

Here is what PDF Accesibility Checker shows:

(The word ’welsh’ is split into two chunks because of a kern. The lang property is inherited by the inner content.)

If you want to look at the PDF code, add the option uncompress to \DocumentMetadata and open the generated file with a text editor. You will see something like this:

26 0 obj
<<  /Type /StructElem /Lang (cy-GB) /S /Span /NS 11 0 R  /P 25 0 R /K [<</Type /MCR /Pg 30 0 R /MCID 1>>  27 0 R ] /ID (ID.07) >>
endobj
27 0 obj
<<  /Type /StructElem /S /Em /NS 11 0 R  /P 26 0 R /K <</Type /MCR /Pg 30 0 R /MCID 2>>  /ID (ID.08) >>
endobj

Note the /Lang (cy-GB). This object contains (inside /K, meaning ‘kids’) the text ‘welsh’ referred to as /MCID 1 followed by the object 27 0 R (the text ‘emphasis’). The marked content for ‘welsh’ is defined before:

/Span<</MCID 1>> BDC
BT
/F15 9.96264 Tf
1 0 0 1 178.63 657.235 Tm [<0072>28<003200480062003F0067>]TJ
ET
EMC

(EMC means ‘end of marked content'. The string is <0072>28<003200480062003F0067> but note it’s not ASCII.)
Here is also the marked content for ‘emphasis’ (/MCID 2):

/Em<</MCID 2>> BDC
BT
/F16 9.96264 Tf
1 0 0 1 205.519 657.235 Tm [<0032004B0054003F001C006200420062>]TJ
ET
EMC

[u-fischer]

\tagstructbegin{tag=Span, lang=id}
      \tl_set:Nn \l__tag_mc_lang_tl {id}
      \tagmcbegin {} ##1 \tagmcend
    \tagstructend

This sets the language in two places: in the structure and in the included mc-chunks. That is not really necessary. One of both options are enough.

Setting the language in a Span structure works fine as long as there is no paragraph break in the argument. \BabelWrapText doesn't allow a paragraph break, but I want to emphasize the point, as \foreignlanguage normally does.

Setting the language on the mc chunks with \tl_set:Nn \l__tag_mc_lang_tl {id} has the advantage that it works across paragraphs and it is less costly as it doesn't create structure elements. But as I wrote in another issue it is currently not completly clear if this is spec conforming. I discussed that at the PDF days and hope for a clarification soon. When this happen I will add an official interface for this method.

Be aware that "works" currently mostly only means "it works in adobe reader on windows" (but foxit will hopefully add support in the near future).
Also to test the output and hear the language switches you need to install the relevant language support with the voices. I typically make examples using german and english and french.

[cfr42]

Given that the maintainers of my OS are largely volunteers working for free, I doubt that would be well received. Besides which, it is really upstream projects that need to do more, which presumably means orca or festival or something and the developers working on pdf-related stuff. (okular, but more likely the stuff okular etc. use. poppler?) and so on.

PDF is really a very problematic format in general, not least because it is so tied to proprietary software. You cannot go to a free software project and say 'this ought to do X and Y because that's what Adobe does'. You could say, 'this ought to do X and Y because that's what this standard says', but, as far as I can figure out, the standards in this case are rather heavily determined by what Adobe does/wants and could not even be freely distributed until fairly recently. That is not the sort of setup which encourages free software implementations.

And it does not much help that TeX, LaTeX et al. are viewed rather sceptically in this context ....

[u-fischer]

They do not need to do "what Adobe does" or use proprietary software. PDF is an open format. You can read and print and edit a PDF on every OS because of this. Open source engines like pdftex or luatex or luametatex can generate since a long time all sorts of PDF because of this. You can not blame Adobe for the lack of accessibility support on your OS. If there is no good support there then it only shows that for the one or the other reason nobody cared enough about it to make it happen.

[cfr42]

Now, yes. But PDF has not been a particularly open format. Even now, the spec is not free, though it is available at no cost. And for a very long time you absolutely could not edit a PDF on every OS.

The lack of accessibility is not entirely Adobe's fault (or, probably by now, even mainly). but they have contributed to the problem.

I doubt it is the case that nobody has cared enough. I think it is much more complicated than that.

[FrankMittelbach]

@cfr42 not sure on what basis you make that claim with respect to Adobe. When PDF was an "Adobe-format" it was open and free: you could get the PDF spec on the web for free (you still can up to 1.7). How do you think Thanh was able to develop pdftex? He did so on the basis a freely available specification (and so could have anybody else, including producing programs to modify existing pdfs). The situation was changed only when Adobe passed the spec on to ISO, because --- don't ask me why --- ISO requires money for any of its standards without exception.

The reason that PDF 2.0 and PDF/UA is now available free of cost is not because ISO changed its policy, it is due to members of the PDF Assocications https://pdfa.org/making-iso-standards-for-pdf-available-at-no-cost/ (one of which is again Adobe) pay ISO so that the standards can be obtained at no costs.

[cfr42]

@FrankMittelbach It is complicated, but the use of non-open extensions in Adobe products (https://web.archive.org/web/20210116133007/https://www.adobe.com/devnet/pdf/pdf_reference.html) makes it hard to develop compatible implementations which can actually deal with all the features available in PDFs produced by Adobe products. So it is not unusual to get a PDF which is only usable in an Adobe reader e.g. the British Library does provide/has provided PDFs of this kind.

Moreover, the way Adobe has licensed technical resources has often been incompatible with free software e.g. https://web.archive.org/web/20160227041714/http://partners.adobe.com/public/developer/support/topic_legal_notices.html. I do not know to what extent these could be avoided, but if you license a patented resource royalty-free, but restrict the scope of use, it cannot be used in free software, even software within that scope. So the whole approach is at odds with any sort of genuine commitment to an open standard.

Adobe wanted PDF to be a standard, but their policies suggest they did not want people to be able to develop compatible software.

Features enabling DRM in PDF are also problematic in this context. This is not solely Adobe's doing. But there are lots of PDFs which you cannot read with anything but an Adobe application and that restriction is quite explicit. If the standard were genuinely open, this might still be true in practice, but it would not be possible to create PDFs which can only be read with Adobe's software. The restriction would have to be to viewers implementing feature X. That might in effect mean Adobe's software, but a genuinely open standard wouldn't enable creators to stipulate a particular implementation, as opposed, say, to a particular feature. But since Adobe has extensions beyond the spec, it is possible to effectively limit people to use of Adobe's software just by using those extensions, which nobody else can implement.

[At least, this has been my understanding. Maybe I am mistaken. Alternatives to Adobe Reader and Acrobat etc. have been on the priority list of free software organisations for years. That doesn't mean people could not have done more. But it does suggest the obstacles here are not simply that 'nobody cares enough'.]

[cfr42]

May I suggest using an example where the language code is not identical with a key word?

[u-fischer]

I agree that id as language is a bit confusing ;-)

[cfr42]

This is setting an internal variable \l__tag_mc_lang_tl. Is this equivalent to setting the language in the argument starting the mc?

[u-fischer]

No. Inside the argument there could be more structures, and they produce their own mc-commands. The \foreignlanguage could produce any number of mc-chunks and you want the lang on every one.

[jbezos]

@cfr42 @u-fischer Thank you for your comments. Rewritten.

[u-fischer]

@jbezos I think you should update the example. With a current babel, this here works quite ok:

\DocumentMetadata{lang=en, tagging=on, pdfstandard=ua-2,uncompress}

\title{Test}

\documentclass{article}

\usepackage[babel-metadata=tagging,german,english]{babel}

\begin{document}

some text \foreignlanguage{german}{mit einem deutschen Zwischentext} \foreignlanguage{french}{et un peu de français} and which continues in english.
Let's try some danish \foreignlanguage{danish} {en sætning på dansk}

\foreignlanguage[actualtext=blub]{german}{mit einem deutschen Zwischentext}
\end{document}

Personally I think it would be better if the tagging is activated automatically if \DocumentMetadata is detected, the extra keys feels a bit complicated (but you could provide a babel-tagging=off to disable).

The actualtext key is probably not so useful in this context. But an useful addition long-term could be to pass on the phoneme key phoneme=[ˈlaːtɛx] (nobody uses it currently, but if we have more examples this could change).

I made some reading tests: for me only adobe switches the voices (for danish it says "Danish not supported and then tries with the english voice). With all other readers only the english voice was used.

[jbezos]

Personally I think it would be better if the tagging is activated automatically if \DocumentMetadata is detected, the extra keys feels a bit complicated (but you could provide a babel-tagging=off to disable).

Sure. Since they are the first steps, for the moment it’s optional, just in case. I’m investigating possible issues related to \DocumentMetadata, which is, in some sense, closely related.

The actualtext key is probably not so useful in this context. But an useful addition long-term could be to pass on the phoneme key phoneme=[ˈlaːtɛx] (nobody uses it currently, but if we have more examples this could change).

I’ve found two approaches. One of them is based on an explanatory text (hence the actualtext, which has, of course, the problem of copy-pasting). Another is /Phoneme (+ /PhoneticAlphabet, although currently it’s /ipa). I’d like to go step by step. And phoneme will be the next step.

Since you’ve confirmed it works, I think this page should be rewritten (not only the example).

[u-fischer]

Actually it is trivial to add language for selectlanguage now too. You only need \tagpdfsetup{text/lang=\BCPdata{language}} (assuming that the \selectlanguage commands are issue e.g. in vmode so that the following structure can pick up the new language.

Or perhaps, to reduce the number of \Lang settings in the pdf, babel should set that only for non-main languages, so something like this

\selectlanguage{german}
\tagpdfsetup{text/lang=\BCPdata{language}}

Dieser Text ist in deutsch

\selectlanguage{english} % main language
\tagpdfsetup{text/lang=} % empty

....

[jbezos]

Although \selectlanguage should be used to change the language between paragraphs, the very fact is many people are doing things like text {\selectlanguage{language} text} text. There was a reason: \foreignlanguage switches neither captions nor date. Well, now there is an optional argument in \foreignlanguage to switch them (since 2020-05-13), but very likely there is a lot of stuff out there with \selectlanguage. I have to analize the impact of adding \DocumentMetadata – perhaps a lot, or perhaps nothing at all, I don’t know. Or perhaps the behavior should be different depending on the mode (although {\selectlanguage{language} text} text at the beginning of a paragraph is not impossible).

[FrankMittelbach]

I would call that insane.

A 人 of strong opinions :-) I don't think I would call Johannes original design that.

[cfr42]

@FrankMittelbach Er ... no. My comment lost a 'not'. It should have read:

I wouldn't call that insane.

I corrected it, but left a note as you replied.

[cfr42]

I do not know what a 人 is.

[FrankMittelbach]

人 = person (in Japanese)

so you are too one of these people who occasionally lose their "not"s in cruical places (just like me)

[cfr42]

indeed, I am afraid so :(. sorry about that. and thank you for the Japanese.