Hesitation Helps

Here's something I didn't know1:

Approximately 6 in every 100 words are affected by disfluency, including repetitions, corrections, and hesitations such as the fillers um and er. Moreover, the distribution of disfluency is not arbitrary. For example, fillers tend to occur before low frequency and unpredictable words, in circumstances where the speaker is faced with multiple semantic or syntactic possibilities, as well as in cases where other types of uncertainty occur. (p. 3)

Well, I knew that I hesitate in speech a lot (my son will often say to me, "Stop saying 'um!'"), but I didn't realize disfluency that common in other people's speech as well. The fact that it is raises all sorts of interesting questions, some of which are answered by the above descriptions of the contexts in which disfluency is most common. Other interesting questions remain, however, the most obvious of which is, how does disfluency affect listeners (other than annoying the hell out of my son)? That's the question that the paper that begins with the above passage, by Corley, McGregor, and Donaldson, attempts to answer

They discuss some evidence that disfluency actually makes it easier to comprehend what someone is saying. For example, in one study2, participants were presented with objects on a computer screen, and given instructions to move the mouse to point to particular objects. The name of the object was either preceded by disfluency (e.g., "Point to theeee umm...."), or not. When it wasn't preceded by disfluency, participants' eye movements showed that they first looked at the object they'd pointed to on the previous trial. However, when it was preceded by disfluency, their eyes went straight to the new object. Based on this and other evidence for a comprehension benefit from disfluency, Corley et al. explored the long-term benefits of disfluency for comprehension in more natural tasks.

To do this, they had participants listen to sentences in which the final world (the target word) was either predictable or unpredictable. For example, participants either heard the sentence, "Everyone's got bad habits and mine is biting my nails," which includes the predictable target word "nails," or "Everyone's got bad habits and mine is biting my tongue, where the target word, "tongue," is unpredictable. In another case, they might hear "That drink's too hot; I've just burnt my nails/tongue." Using each target word as a predictable and unpredictable target allowed for better control. For each sentence, the target word was either preceded by disfluency (the word, "er") or not.

If disfluency aids comprehension, then people should have an easier time comprehending sentences ending with unpredictable target words if those words are preceded by "er." To measure comprehension, Corley et al. used two dependent variables. First, they measured event related potentials (ERPs), obtained with electrodes placed on participants' scalps. ERPs are commonly used in language comprehension studies because of something called the N400 effect. About 400 ms after a listener hears or reads the beginning of a word, the N400 component of ERPs over certain language-related areas of the brain peaks if the person is comprehending the word. So, the presence of the N400 effect for target words would indicate that people are understanding those words easily. As a second measure, participants were given a memory task. They received all of the target words from the first task (160 in all), along with an equal number of new words (foils), and were asked to indicate which of the words they had heard in the previous task.

Looking first at the ERP data, they found that for fluent sentences (sentences in which the target word was not preceded by "er"), predictable target words showed more N400 ERP components than unpredictable targets. For sentences in which target words were preceded by hesitation ("er"), there was no difference between the predictable and unpredictable words. The disfluency apparently made it easier to comprehend the unpredictable target words. In the memory task, unpredictable target words were better remembered, overall, than predictable words. This isn't surprising, given that unexpected information is generally better remembered than expected information. For predictable target words, however, memory was better if the word had been preceded by "er" than if it had not.

Combined, these results indicate that comprehension (and as a result, memory) is better for both predictable and unpredictable words when they're preceded by disfluency. It's not really clear from the results of this and previous studies why, exactly, disfluency aids comprehension. Intuitively, it seems as though hesitation would cause us to further process the context of the sentence, making unpredictable words more difficult to comprehend after hesitation. However, this obviously isn't the case. Corley et al. speculate that the hesitation may actually allow people to distance their processing from the context, making words that are unpredictable from the context less difficult, but that explanation will have to be tested in further studies. For now, however, it's clear that hesitation can be an important rhetorical device. If you want people to understand difficult utterances, in which they may have to comprehend novel information, disfluency may be important. Hesitating before uttering words that you want your audience to remember may be effective as well. I'd also be interested in seeing the effects of disfluency in comedy. Comedians and comedy actors all know the importance of timing in comedy, and since one of the important aspects of humor is the unexpected (unexpected punch lines are much funnier than expected ones, for example), I wonder how hesitation before the punch line effects its perceived humor.


1Corley, M., MacGregor, L.J., & Donaldson, D. I. (In press). It's the way that you, er, say it: Hesitations in speech affect language comprehension. Cognition.
2Tanenhaus, M.K., Spivey-Knowlton, M.J., Eberhard, K.M., & Sedivy, J.C. (1995). Integration of visual and linguistic information in spoken language comprehension. Science, 268, 1632-1634.

More like this

Sometimes I forget that not everyone who happens upon this blog today has been reading it from day one (I mean come on, why haven't you?). It surprises me, then, when people tell me they've seen no evidence that George Lakoff and Mark Johnson's conceptual metaphor theory is, well, wrong. I guess I…
Trueswell & Kim's paper in the Journal of Memory and Language describes a phenomenon known as "fast priming," in which a reading task is momentarily interrupted by a brief presentation of a "prime" word, usually lasting around 30 to 40 ms. The reading task then continues, and although subjects…
There are 9 new articles in PLoS ONE today. As always, you should rate the articles, post notes and comments and send trackbacks when you blog about the papers. You can now also easily place articles on various social services (CiteULike, Mendeley, Connotea, Stumbleupon, Facebook and Digg) with…
THINKING of and saying a word is something that most of us do effortlessly many times a day. This involves a number of steps - we must select the appropriate word, decide on the proper tense, and also pronounce it correctly. The neural computations underlying these tasks are highly complex, and…

My guess would be that disfluency is like a pointer to a new word.

40 years ago, the term was 'nonverbal pauses', and the people in charge assumed they had no useful function -- despite endless anecdotal evidence to the contrary. They devised techniques to train people not to make these NVPs, but very little money ever got coughed up for it, and there was no clinical evidence the training improved anything measurably.

We are so good at noticing slipups -- say a reformed stutterer hesitating every so slightly in getting just one word out -- yet the 'nonverbal pauses' usually slip right past us. That told us something but nobody was listening.

I see you cited the Tanenhaus et al. (1995) paper. Have you seen any Julie Sedivy's follow-up research? She has been investigating similar questions about how prenominal adjectival modifiers influence eye-tracking measures of reference determination in a visual array. Her research has turned up good evidence for Gricean-like processes in integrating adjectives to signal contrast. This disfluency research seems to admit of a similar processing story.

When I was in grad school and we were learning to speak in public (first, in front of each other), part of the training was to become aware of such disfluency so as to overcome it. Now I think I understand better why my sometime students think I speak too - densely. I'll think a bit more about timing the next time I speak in public (tomorrow, ouch).

A lay explanation for the increased comprehension may have to do with empathy? I know that when a speaker is obviously searching for a word, I internally try to fill it in for them and show them that I get the gist with a nod or a look, so they can move on. Sometimes listeners even say the word to "help" the speaker, perhaps through impatience to get to the next new concept? Disfluency and pauses might incite the listener to make that extra effort toward the speaker so as to get past that point in the sentence, and when the word is difficult or unknown that effort translates into better comprehension on the part of the listener.