The oddities of the bigram "ed"

[Dunsel]

Culpepers total rk occurrences: 379/2 = 189 & 190. And that's 1321 words per page.  76%

Seems right.  My version has N=222 occurrences of "rk", hence my suggestion would be N/2 = 111 pages.  My version has 123832 words, so it would be 1115 words/page.  

It is hard to tell whether "rk" is anomalous in your plot because all the rare digraphs, including "rk", are squeezed into the few mm near the left margin.

Maybe the graphs would be more legible if the vertical axis was changed to the percentage of pages without the digraph, and both axes were in log scale.  Then I guess that you would get a more or less straight line, similar to the Zipf plot in log-log scale.

If you say that "rk" occurs in only 76% of the pages, instead of the expected 86%, that would already indicate uneven distribution.  Not as dramatic as ed, but it would show how topic can cause the uneven distribution of certain digraphs.

All the best, --stolfi

You are likely right that changing the scale would blur things, I'm certain.  All of those scatter charts are a simple percentage on the y axis, count on the x axis.  I'll kindly ask you to wait for my third planned post and then, I believe you'll see your 'rk' analogy is nothing like ed.

---

It is easy to explain if we assume that the text is gibberish. One scribe liked to use it while other didnt. And Currier languages correspond with supposed different hands. That would be all.
Meaningful explanation is harder. Writing in another plain language would probably change more text properties, not only this

It's not "only this". There are other bigram statistics (like "lk", "ll", "lr"), and other properties that Currier noticed...

Check pt 2 of this series.  You'll see that it's not just those.  60 new bigrams get added in the ed+ pages.