spoken

Gather a recogniser's tokens into the words they spell.

A recogniser answers in tokens, not in words: be, aut, y's. Where one word ends and the next begins is the recogniser's convention, not the caller's, and every side that reads the same model has to read that convention the same way — otherwise one returns beauty's where another returns beautys and a third splits the word in two, and the three cannot be held against each other however alike they heard the sound.

Each word is timed from the token it opens with to the one it closes with. Marks at either end of a word are left off: a word the model wrote increase, is the word increase, and the comma is the model's punctuation rather than anything it heard. Marks inside a word stay, because beauty's and self-substantial are words and beautys is not.