XeTeX ⇒ Ligatures and search

[paul]

After typesetting a book using Xe(La)TeX, I was asked why some "character pairs get squeezed into strange binary single characters in the PDF" -- i.e., ligatures for "fi", etc. After explaining, I was asked if it was possible to have the ligatures appear in the "visual layer" and the individual component letters in the "text layer" (I didn't even know there was such a thing!), because the ligatures interfere with searchability.

Anybody have a solution?

[josephwright]

This should usually "just happen". Can you provide an example where you are seeing an issue.

[paul]

From my testing: evince recognizes ff/fi/fl/ffi/ffl ligatures, but no others as far as I can tell (the other common one being Th). xpdf doesn't recognize any of them. I haven't tried Adobe software, but the publisher tells me it's a problem.

[josephwright]

N, I meant can you provide the TeX source for something showing the issue, so we can test it ourselves.

[paul]

Sure; but anything that uses the "problem" characters will do.

Code: Select all

Code, edit and compile here:

1
2
3
4
5
6
7
\documentclass[a4paper]{article}
\usepackage{xltxtra}
\setmainfont[Mapping=tex-text]{Adobe Garamond Pro}
\begin{document}
This is a problem.
\end{document}
 
 
הההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההה
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Run LaTeX here

[frabjous]

This seems to be very font specific. I tested with this:

Code: Select all

Code, edit and compile here:

1
2
3
4
5
6
\documentclass{article}
\usepackage{fontspec}
\setmainfont[Mapping=text-tex]{Adobe Caslon Pro}
\begin{document}
The affluent affinity for fishing off the flank of the fjord \ldots
\end{document}
 
 
הההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההההה
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

Run LaTeX here

Evince could "see through" (when searching) the ffl and ffi ligatures, but not the Th or fi or fl or fj ligatures.

I then switched to Sorts Mill Goudy and evince couldn't see any of the ligatures.

(I chose Sorts Mill Goudy since it's one of the few freely available opentype fonts out there that has ligatures. I figured some people reading this might want to test for themselves even if they don't have access to the expensive Pro fonts. It does not have a Th-ligature though.)

(P.S. Testing some more, Okular seems to have more problems than evince, so this is viewer-specific too. )

[paul]

Also, if you have old-style digits, you can't search for "normal" digits.

[paul]

OK, so what's happening is that when the ligature exists in Unicode (that is: IJ, ij, LJ, lj, ff, fi, fl, ffi, ffl, st -- but not OE or oe), everything works OK. Similarly for the Unicode double-letters (somehow distinguished from ligatures by Unicode naming) LJ, Lj, lj, NJ, Nj, nj, DZ, Dz, dz, DŽ, Dž, dž -- but, again, not AE or ae. All other printed ligatures are "broken" because they're at some private-use codepoint lacking Unicode decomposition data -- thus the same problem occurs with alternate forms.

Is it possible to invisibly associate the components with the ligated characters in the same way OCR text can be associated with scanned documents?