5 ms·

Let's take an example: https://arxiv.org/pdf/1606.06389.pdf https://arxiv.org/pdf/1606.06389.pdf A cursory inspection seems to indicate ctrl-F is fine here?

by marvy 6y ago

Let's take an example: https://arxiv.org/pdf/1606.06389.pdf https://arxiv.org/pdf/1606.06389.pdf A cursory inspection seems to indicate ctrl-F is fine here? Or am I missing something?

j-pb 6y ago
Single column layouts often work semi-acceptably for text only, when not purposefully tampered with. 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 ^^?
- marvy 6y ago
  No need to go on, that's already more than I bargained for, thank you so much for taking the time to respond :) 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?
  - j-pb 6y ago
    I was using the builtin viewer of chrome I think, could also have been safari. 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.
    - marvy 6y ago
      I just tried in Chrome; at least for page 2 it actually did a bit better than my browser: it actually managed to preserve the line breaks! (I don't have Safari installed.) (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.)
lokedhs 6y ago
Try copying across a line break. You'll notice that the spaces between the words are missing when you paste it.
- marvy 6y ago
  Just tried copying across a line break on page 9, seems to work: Here the potential function is, to say the least, not very simple.

by marvy 6y ago

Let's take an example: https://arxiv.org/pdf/1606.06389.pdf https://arxiv.org/pdf/1606.06389.pdf A cursory inspection seems to indicate ctrl-F is fine here? Or am I missing something?

j-pb 6y ago

Single column layouts often work semi-acceptably for text only, when not purposefully tampered with. 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 ^^?

marvy 6y ago
No need to go on, that's already more than I bargained for, thank you so much for taking the time to respond :) 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?
- j-pb 6y ago
  I was using the builtin viewer of chrome I think, could also have been safari. 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.
  - marvy 6y ago
    I just tried in Chrome; at least for page 2 it actually did a bit better than my browser: it actually managed to preserve the line breaks! (I don't have Safari installed.) (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.)

lokedhs 6y ago

Try copying across a line break. You'll notice that the spaces between the words are missing when you paste it.

marvy 6y ago
Just tried copying across a line break on page 9, seems to work: Here the potential function is, to say the least, not very simple.