Issue with sampling (especially with verses, and in combination with n-grams)

With current version of random sampling of words or verses, features such as n-grams that would be over several units risk showing combinations that do not exist in the data…