Sentence level translations, rather than word or paragraph level, graded by difficulty as beginner/intermediate/advanced.

You can look up more about that research from here: https://en.m.wikipedia.org/wiki/Extensive_reading

Ah, ok, now I understand. The problem might be that we aren't using machine translated texts, but human translated. So when matching translations, often sentences don't have a 1-to-1 matching. Translators often change sentence structure to better suite the new language (there are even instances of translators adding new characters in the book).

In any case, it's a very interesting problem