When you ask people to judge others by their speech, a trend emerges: Listeners dislike disfluency. Slow talkers producing loads of ums and pauses are generally considered as less charming. But science tells us there may be even more to disfluency.
Disfluencies do not occur in arbitrary positions in sentences. Ums typically occur right before more difficult or low-frequency words. Imagine you're having dinner with a friend at a restaurant, and there're three items on the table: a knife, a glass, and a wine decanter (醒酒器). Your friend turns to you and says, "Could you hand me the…um..." What would you assume your friend wants? Since it's unlikely that they will hesitate before such common words as knife, and glass, chances are you'll pick up the decanter and ask, "You mean this?"
This is exactly what we demonstrated through controlled eye-tracking studies in our lab. Apparently, listeners hear the um and predict that an uncommon word is most likely to follow. Such predictions, though, reflect more than just simple association between disfluencies and difficult words; listeners are actively considering from the speaker's point of view. For example, when hearing a non-native speaker say the same sentence but with a thick foreign accent, listeners don't show a preference for looking at low-frequency objects. This is probably because listeners assume non-native speakers may have as much trouble coming up with the English word for a common object, like a knife, as for unusual ones and can't guess their intention.
In another experiment, listeners were presented with an atypical speaker who produced disfluencies before simple words and never before difficult words. Initially, participants displayed the natural predictive strategy: looking at uncommon objects. However, as more time went by, and they gained experience with this atypical distribution of disfluencies, listeners started to demonstrate the contrary predictive behavior: They tended to look at simple objects when hearing the speaker say um.
These findings represent further evidence that the human brain is a prediction machine: We continuously try to predict what will happen next, even though not all disfluencies are created equal.
