June 6, 2002 -- So, there are these, uh, psychologists who believe that um, I mean, they're trying to prove that those little stops and starts that are, you know, used in conversational speech are, like, legitimate words.
Herbert Clark of Stanford University and Jean Fox Tree of the University of California at Santa Cruz have spent years listening to recordings of spontaneous conversations and speech to analyze the role of "ums" and "uhs" in language.
And unlike previous linguists, they've concluded these so-called disfluencies and discourse markers represent something more than clumsy speakers having trouble expressing themselves — they also serve a role for listeners.
"People use these phrases in a very particular, deliberate way," says Clark. "If we anticipate a delay in our speech, we choose the appropriate sound to signal this to the listener. These phrases mean 'I need to make sure you realize I'm delaying because I'm having trouble.'"
By signaling a delay is coming, a speaker avoids a silent gap in conversation that might otherwise prove confusing to a listener.
"When we talk, we have to do two things," says Clark. "We have to pay attention to the content of what we're saying and also keep track of the interaction of two people talking."
Phrases like "um" and "uh" and "you know" play an important role in language, he argues, by serving as a speaker's "conversation managers" in the human interaction aspect of conversation.
That idea runs counter to the thinking of Noam Chomsky, a renowned linguist at the Massachusetts Institute of Technology who argued in the mid-1960s that such phrases are simply "errors in applying knowledge of language in actual performance." Chomsky didn't consider "um" and "uh" part of proper language and he influenced a generation of linguists to exclude such phrases from linguistic theory.
Um vs. Uh
Clark and Fox Tree are working to change that tide by proving these phrases play specific roles in conversation. Their analyses show that the "um" sound almost always sets up a long delay in speech, while the sound "uh" signals only a brief pause.
Speakers also extend the vowels of words for the same effect. When people have trouble finishing a sentence, for example, they're more likely to extend the vowel in "the" to sound like "thee" as they search for the next word.
Occasionally becoming stuck mid-sentence is natural, argues Clark, considering that people speak 120 to 150 words per minute — or two to 2 ½ words per second — during normal conversation. At that rate, it's only natural to experience a few stops and starts.
Studies show that disfluencies make up an average of 6 percent to 10 percent of spontaneous speech. And all languages have them. The French say "eu" and "em," Spanish speakers say "eh" and "pues" and Japanese say "etto" and "ano" to name a few.
Like, You Know
Fox Tree is also tuning into English phrases such as "you know," "I mean," "oh" and "like." Her early studies suggest they're often uttered to adjust the meaning or structure of a sentence as it's being spoken.
For example, when a person says "I need to pick up orange juice and milk today, oh and cheese," the "oh" suggests new information is being added that actually belonged earlier in the sentence. When someone says "She would be a good runner, I mean swimmer," the phrase "I mean" is used to adjust or correct something that was said earlier.
She says she's still a little stumped by the use of "like" in speech since it's used so many different ways. But, she argues, that doesn't mean it isn't legitimate.
"In school we were taught not to use these phrases, but the fact is, everybody uses them," said Fox Tree, who says her high school debate teachers used to discourage the use of "um's" and "uh's." "So we think they must have a role."
Computers That Can Go With the Flow
Another group of researchers paying close attention to how we speak are those working on developing advanced language recognition systems that help computers understand human speech. James Glass, a professor of spoken language systems at MIT, says he and his team have created models for phrases like "uh," "um" and "I mean" so that voice recognition systems will know their sound — and then tune them out.
"If you're not expecting to hear them, it can cause an error to occur," he said. "It improves our accuracy to recognize these speech effects that are a natural part of spoken language."
Although linguists and computer scientists may be working to elevate the significance of sounds like "um" and "uh" in everyday language, Fox Tree admits that, to some degree, her high school debate instructors were right.
"It's true if you use these words too much it can be distracting," she says. "Then you're constantly forcing the listener to try and finish your thoughts."