Hey DCH, it occurs to me that one way of usefully constraining this would be to capture the pos analysis of the supply text in a calculated constant before the first iteration and then feed the resultant elems of this array into the any similar-word methods. Why? to avoid any syntax drift cause by puns with more than one pos. i'm gonna do this for my quatrains fork

indeed -- you can see this in the paravec(https://github.com/dhowe/paravec/blob/main/data/words.js) iteration, which pre-captures a bunch of info