I could be misglancing this at 1 am, but it looks like a bit of pomp to me. The only data they have are the lengths of the translations (in syllables) and how long it took to read them. The "information density" is just a fancy way to report the former, and they only reveal the latter through the rates. It seems they simply observed that 1. translations differ in length but 2. are read in about the same amount of time. Alas, talk of "information density" had me hopeful they applied a protocol like Shannon's to the spoken domain.
http://languagelog.ldc.upenn.edu/myl/Shannon1950.pdf