We, uh, signal speech pauses

Sounds differ among cultures, but practice is, um, universal

Michael Erard
New York Times
Jan. 11, 2004 12:00 AM

If you were hearing this instead of reading it, you might notice a pause here and there tucked between the phrases, filled with a familiar, soft hum or rumble - an um or uh.

Though a bane to teachers of public speaking, people around the world fill pauses in their own languages as naturally as watermelons have seeds. In Britain, they say uh but spell it er, just as they pronounce er in butter.

The French say something that sounds like euh, and Hebrew speakers say ehhh. Serbs and Croats say ovay, and the Turks say mmmmm.

Japanese say eto (eh-to) and ano (ah-no); the Spanish, este; and Mandarin speakers, neige (NEH-guh) and jiege (JEH-guh).

In Dutch and German, you can say uh, um, mmm. In Swedish, it's eh, ah, aah, m, mm, hmm, ooh, a and oh; in Norwegian, e, eh, m and hm.

These interruptions, it turns out, plague machines more than people - speech-recognition systems in particular. Researchers, therefore, increasingly have been turning their attention to uh and um (among other so-called disfluencies).

"If someday you want machines to be as smart as people, then you have to have machines that understand speech that's natural, and natural speech has lots of disfluencies in it," said Liz Shriberg, a research psychologist at SRI International, a research company based in Menlo Park, Calif.

Uh and um may tell a computer about a speaker's alertness or emotional state so the system can adjust itself and let people speak naturally to speech-to-text programs.

Studying disfluencies

Well before the invention of speech recognition technology, Frieda Goldman-Eisler, a psychologist in London in the 1950s, inaugurated the modern study of disfluencies by developing instruments that counted pauses in speech and measured their duration.

Goldman-Eisler, who was looking for a way to make psychiatric interviews more efficient, found that 50 percent of a person's speaking time is made up of silence. She also hypothesized that a speaker planned his next words during the length of the uh or um.

Around the same time, a psychiatrist at Yale, George Mahl, counted uhs and nine other speech disfluencies in order to measure a person's anxiety level, calculating that during every 4.4 seconds of spontaneous speech, on average, one disfluency occurs; 85 percent were uh and um, restarted sentences and repeated words.

A slip of the tongue, upon which Sigmund Freud practically built an intellectual career, occurred less than 1 percent of the time.

Goldman-Eisler and Mahl treated uh and um as symptoms of nervousness and verbal struggle. Once cheap, fast computers made digitized speech easy to study in the 1990s, the approach changed. Researchers began to study verbal pauses for meaning; they focused on the words as information.

It's, um, a signal

By far the newest, and most controversial, idea comes from Herbert Clark, a psychologist at Stanford University, and Jean Fox Tree, a psychologist at the University of California-Santa Cruz, who determined that speakers use (and listeners understand) uh and um in distinct ways.

Uh signals a forthcoming pause that will be short, while um signals a longer pause, she said. Uh and um are not acoustic accidents, but full-fledged words that signal a delay yet to come.

Fox Tree, who is a former student of Clark's, became interested in uh and um as an undergraduate majoring in linguistics at Harvard. There, she realized that theories of language could not account for the fragmented nature of ordinary conversation.

"I thought, here's something you hear in every single conversation during the day, some kind of disfluency, and yet people treat them as if they're garbage," she said. "Why are they there? Why do we use them?"

Fox Tree studies other discourse markers like you know, I mean and oh, and is working on so and and. Her dream topic is like.

"I waited before I got tenure to study like," she confessed, "because I thought it was going to be messy and hard to get a hold of, and I would spend all this time studying it."

Shriberg agrees that these disruptions are more than white noise.

"When you realize these things are distributed in very clean ways and have a very elegant structure," she said, "then you can see they're not garbage at all."

Heather Bortfeld, a psychologist who studies infant language development at Texas A&M University, discovered this through personal experience. While living in Madrid during her junior year in college, she noticed the distinct sounds the Spanish used to fill their pauses.

"These were often conveying important information that I had to learn about," Bortfeld said.

"Then I had to learn how to make them myself in order to sound more native and really to be speaking Spanish correctly."

In 2001 Bortfeld and others reported in the journal Language and Speech that speakers taking a more active role in tasks said uh and um, repeated words and restarted sentences more frequently than those in a passive role.

Men say uh and um more than women, though their overall disfluency rate was the same. One piece of conventional wisdom fell by the wayside: Whether the speaker and listener knew each other had no effect on rates of uh or um.

Still, it may be Nicholas Christenfeld, a psychologist at the University of California-San Diego, and other researchers who have come up with the most appealing findings.

He counted uh among professors giving lectures and found that the humanities professors say you know and uh 4.85 times per minute, social scientists 3.84 and natural science professors 1.39 times. He said it suggests that humanists have more expressive options from which to choose.