Let's take an example:
https://arxiv.org/pdf/1606.06389.pdf
A cursory inspection seems to indicate ctrl-F is fine here? Or am I missing something?
https://arxiv.org/pdf/1606.06389.pdf
A cursory inspection seems to indicate ctrl-F is fine here? Or am I missing something?
However if you try to select the text on page 2, you'll see that the layout engine didn't actually place the formula into its own section. Not only is it broken, with the sum signs missing, it is also jumbled up with the text below it, resulting in weird - text broken formula - text - broken formula - mumbo jumbo.
This would make it extremely hard if not impossible to properly follow the paper with a screen reader.
The figures on page 6 also produce weird artefacts on the text layer. In HTML for example you'd simply present the alt-text to the person with the screen reader, but TeX doesn't have such a feature in most vector drawing libs afaik.
The table on page 8 is also not correct in terms of the text layer. If you select it you can see that you get the top half of the leftmost column followed by the rightmost column, followed by the rest in some order.
If I may also direct your attention to this gem on page 9: "Then at least half of allnodeshaveh ̄=lgn =lgn−1(seethepaperforthedefinitionofh ̄),and 2√ thus a potential of Ω( lg n). Therefore, the total heap potential would be lg n). Conjecture: the same construction works for the fancy potential √ √ function, which would give a bound of Θ(n · 4 lg lg n)."
Shall I go on ^^?
I suspect we're using different PDF viewers; I'm using the one that comes with my browser.
The results are somewhat less disastrous than you describe, but still bad. (I'm seeing maybe half the problems you mentioned.) I'm a bit curious which viewer you're using.
I can describe what happens when I copy/paste the stuff you mentioned on pages 2/8/etc., and if you're interested I will, but if you're not interested let me ask a slightly different question:
Rather than try to get the screen reader to make sense of the final PDF, would it be easier to just download the original page source from arxiv and let the screen reader deal with that?
Using the original TeX source for the screen reader is significantly hindered by the fact that TeX isn't a markup-, but a programming language.
TeX is turing-complete by design, making it infinitely extensible, in order to avoid knuth ever having to re-typeset his books. After all if it's a programming language and a new system, problem, style, e.t.c, comes along you can just write a program that deals with it.
But this has horrible effects on render-ability, in order to know what the final document should look like, you need to run it, no way around that, thanks to Rice's theorem.
LaTeX users also generate a lot of their figures with TeX itself, write their own styling or bibliography rules, and write their own custom graphics rendering libraries.
The easiest path is to just give up, render the entire PDF to a 300dpi lossless image, and throw it into an OCR engine. These things contain a buttload of heuristics to generate structure and meaningful text based on visuals, and since we know that humans explicitly (and often only) care about those in TeX documents...
It's pretty darn sad, because in many ways TeX is holding scientific advances back, by eating its own children. My guess is that if scientific papers had branched off of plaintext tools like troff, we'd probably publish papers as machine readable semantically annotated knowledge-graphs by now.
(Also, if you look closely, the summation signs are not gone, they are replaced by the letter P. Which is not helpful I admit.)
Do you think there's any place here for education/advocacy? For instance, everyone who makes web pages knows to provide alt text for images.
If there was a standard package that everyone knew they had to include or else it breaks everything from ctrl-F to copy/paste to screen readers, presumably people would use it, right?
I'm less interested in speculating what would have been if troff had "won", (though it is indeed fun to speculate), and more interested in how to fix the mess we're in now, so that 10 years in the future, blind people have better choices than OCR.
(Though OCR is still an improvement over the best option in the 1980s I bet. Though I wasn't around so just guessing.)
That's not a culture thing, this is by the mechanism of <img alt. You won't get people who write TeX to change all of their workflows, and even if there was such a culture.
"If there was a standard package that everyone knew they had to include or else it breaks everything from ctrl-F to copy/paste to screen readers, presumably people would use it, right?"
Then it would still be nigh impossible because TeX commands, like all programming languages, compose rather poorly. It would be a herculean effort to produce a kinda but not really TeX that is both accessible with a focus on semantics, yet still compatible with the billions of lines of LaTeX/TeX out there.
"I'm less interested in speculating what would have been if troff had "won", (though it is indeed fun to speculate), and more interested in how to fix the mess we're in now, so that 10 years in the future, blind people have better choices than OCR."
Boycott LaTeX/TeX and PDF everywhere you can. Whenever you publish a paper, also publish it in markdown/html. Publish in OpenAccess Journals like [PeerJ](https://peerj.com/) which convert all of their papers to html in addition to pdf. Consider publishing papers in alternative forms like nextjournal.com .
We need to get our priorities straight in academia :/. This entire "but latex produces such beautiful documents", "I'm working towards getting into the most prestigious journal" culture of snobbery and vanity needs to stop. We need to go back to caring about the content, not the presentation, something TeX ironically was meant to do.
Fine, fair enough. If fixing TeX is hopeless, then so be it. I assumed it just needed a few small tweaks, maybe combined with slightly cleverer PDF viewers. Guess I was wrong.
But then what should people use for math? I suppose there's MathJax, which seems to have put some thought into accessibility.
There's still a problem though. I can't help but notice that the journal you linked to is a biology journal. In some math/CS circles which are TeX's "home turf", TeX is far more entrenched to the point where I'm not sure such things even exist. For instance, arxiv sort of supports HTML, but not really:
https://arxiv.org/help/submit_html
So there I guess step 1 is to make HTML a viable option.
I think MathJax is certainly a step in the right direction, they even support rendering to MathML.
But I agree that there is a certain lack there in terms of full semantic representations. MathJax is more accessible than TeX but it's still describing visual layout, instead of semantic meaning.
Pushing HTML to arxiv is also a step into the right direction.
I think the most important thing we can do is not be complacent with the state of the art. We need to go back to an age of computing where we didn't think we had it all figured out. We need to experiment, and not be afraid to take a step back in some aspects, like layout and kerning, in exchange for other advances like semantic representations and knowledge representation.
I think bred victor has a great talk on this: https://www.youtube.com/watch?v=8pTEmbeENF4
I think we need to experiment with things like observablehq.com or nextjournal.com or the many other that are coming into existence.
re: semantics vs visual layout of math... Wikipedia says OpenMath is a thing, but... that only solves half the problem. Once you have a format that encodes what you want, someone has to actually it.
Like, if some writes x^{-1} and f^{-1}, it's hard for a computer to figure out that the first one means "the number you get when you divide 1 by x", whereas the second one means "the function you get when you compute the inverse of f".
And if the author can't be bothered to slow down and say which is which, then the reader will have to guess.
re: HTML to arxiv: not ready for prime time, if you actually follow that link.
re: kerning: TeX's advantage here is not fundamental, I think. Just need a good font, as far as I know. (Actually that's not far; I know almost nothing here.)
re: layout: CSS is finally getting good at this from what I hear.
re: talk: looks familiar; maybe I should re-watch it.
Here the potential function is, to say the least, not very simple.