May 10, 2002

Study finds speakers overestimate clarity

A study by a pair of University of Chicago psychologists has found that speakers systematically overestimate their ability to convey what they mean, even when they know that what they say is ambiguous. In contrast, an overhearer who knows a speaker's intended meaning does not systematically overestimate the speakers effectiveness at conveying that meaning. The team suggests that "the very attempt to communicate contains the seeds of miscommunication."

These findings were derived from an experiment conducted by Boaz Keysar and Anne S. Henly of the Social Sciences Collegiate Division, and their research article appears in this month's issue of Psychological Science, the journal of the American Psychological Society.

"While previous studies have shown that people are often misunderstood, this study is the first time it has been examined whether people are aware of their effectiveness at communication," said Keysar, who is an associate professor of psychology. "Our study reveals that we tend to have no appreciation that we don't do a good job."

The experiment consisted of two parts, one involving speakers and listeners and the other involving overhearers. In the first part, 40 pairs of participants were given a set of 12 syntactically ambiguous sentences and four lexically ambiguous sentences. For each sentence, the person acting as the speaker had to convey a specified meaning, while the listener had to guess what the speaker meant.

An example of a syntactically ambiguous sentence used was "Angela killed the man with the gun," where either Angela or the man had the gun. A lexically ambiguous sentence was "Currency is no longer exchanged by the banks, "where the banks could mean either financial institutions or riverbanks.

Speakers then estimated how successful they had been at being understood, and listeners estimated how successful they had been at understanding. While participants sat facing each other as the speaker read a sentence, after each sentence the speaker would immediately turn around, "so that visual cues from facial expressions could not influence speakers' estimations of effectiveness or listeners' confidence ratings," Henly said.

Participants were mostly University undergraduates, and all were native speakers of English. Keysar and Henly did not systematically manipulate whether participants knew each other. "It is unlikely that they did as they were scheduled independently," Henly said.

They found that 72 percent of the time, speakers believed that their intended meaning was being understood, when 46 percent of the time they thought they were understood they were actually misunderstood. The fact that speakers predominately overestimate their effectiveness indicates that errors in estimation are systematic rather than random, as random errors would result in similar rates of over- and underestimation.

The listeners misunderstood the speakers 39 percent of the time. However, when listeners were confident that they understood, they were just as likely to have misunderstood, indicating that the listeners' confidence was unrelated to their accuracy." This suggests that feedback to the speaker might have little diagnostic value, and might provide the speaker with few reliable opportunities to learn how ambiguous his or her utterances are," Keysar and Henly said in the article.

In the second part of the experiment, 37 participants, who had been told what the speakers were trying to convey, tried to guess how effective the speakers had been at communicating to the listeners. They were not present during the speaker-listener sessions, but listened to recordings of the speakers reading the sentences.

These overhearers thought that the speakers were understood 56 percent of the time, in contrast to the 72 percent rating the speakers gave themselves. "The fact that the overhearers, unlike the speakers, do not have a tendency to overestimate, indicates that it is the act of speaking that is responsible for the overestimation," Keysar said.

The team suggests that speakers overestimate their own effectiveness because of the mental effort involved in speaking, as well as the speaker's knowledge of methods he used to disambiguate his speech." Speaking requires more cognitive effort than overhearing, and this use of mental resources could be why speakers were unable to estimate as accurately as the overhearers," Keysar said.

He also pointed out that a speaker is aware of verbal cues, such as pauses or stresses, which he or she has used to convey the meaning of a sentence. "This privileged knowledge could bias one towards thinking that one has managed to strongly convey meaning through intonation, when the fact is that one was still not clear enough," Keysar said.

This tendency to overestimate also correlates with previous studies where people were found to overestimate their ability to control chance outcomes via their actions. "This illusion of control can be seen in the observation that people prefer to use lottery tickets for which they have selected the numbers, even when they are given the choice to switch to tickets with better odds of winning, indicating their belief that their selection of the numbers increases their chances of winning," Keysar said. "In contrast, observers did not overestimate these people's chances of winning, just like our overhearers did not overestimate our speakers' effectiveness."

Keysar and Henly recognize that a potential drawback to their study is that the speakers were given sentences to read instead of making their own sentences, and that perhaps the kind of ambiguity presented in the experiment's sentences are largely avoided in natural conversation. However, they note it is highly unlikely that speakers naturally avoid syntactically or lexically ambiguous sentences and suggest that in cases of self-constructed sentences, speakers would be even more likely to overestimate their success at conveying its meaning than they would in the case of a sentence given to them.

"In the experiment, the speakers are aware that all the sentences are ambiguous, and so they are primed to the possibility of being misunderstood," Keysar said. "But in everyday interaction, people don't appreciate how ambiguous language is in general, and so we don't realize how often what we say can be misunderstood."

While Keysar does not offer a hard-and-fast rule on how to avoid ambiguity in verbal communication, he cites an awareness of potential ambiguity as a starting point. "We have to realize our natural tendency towards ambiguity. I don't know if it will be a great help in reducing ambiguity, but it is a good first step," he said.