“Experiments with ums or uhs spliced in or out of speech show that when words are preceded by disfluencies, listeners recognize them faster and remember them more accurately. In some cases, disfluencies allow listeners to make useful predictions about what they’re about to hear.”