Look at *differences* between words rather than at the words themselves?

[radapox]

Are you sure that's the case though? TTR doesn't look "inside" words, so even a change of one letter is a change. Since your method inevitably cycles around based on the previous letters, my intuitive expectation is that no matter how you divide your codes, TTR will be unusually high. But this is difficult to predict intuitively.

Hah, good point there! Yes, of course, you're right. Different is different, it doesn't matter how different.