New research uses computer modeling to understand how people communicate with nonliteral language, such as metaphors, hyperboles, and exaggerated statements.
The findings show that people understand nonliteral language when they realize the purpose of the communication.
Noah Goodman, an assistant professor of psychology at Stanford, believes that figurative language—the nuanced ways that people use language to communicate meanings different than the literal meaning of their words—is one of the deepest mysteries of human communication.
“Human communication,” he says, “is rife with nonliteral language that includes metaphor, irony, and hyperbole. When we say ‘Juliet is the sun’ or ‘That watch cost a million dollars,’ listeners read through the direct meanings—which are often false if taken literally—to understand subtle connotations.”
Sharp and round numbers
In the Computation and Cognition Lab, the researchers developed computational models that use pragmatic reasoning to interpret metaphorical utterances. Their research for this particular project involved four online experiments with 340 subjects.
Participants in the experiments read different scenarios involving hyperbole. For example, a person bought a watch and was asked by a friend whether it was expensive. That person responded with different figures, ranging from low- to high-cost figures – such as $50, $51, $10,000, or $10,001.
Given this, the participants rated the probability of the purchaser thinking it was an expensive watch or not.
People tended to interpret “sharp numbers”—such as a watch costing $51—more precisely than “round numbers,” as in a watch costing $50.
The findings, published in the Proceedings of the National Academy of Sciences, suggest that even creative and figurative language may follow predictable and rational principles.
“This research advances our understanding of communication by providing evidence that reasoning about a speaker’s goals is critical for understanding nonliteral language,” says first author Justine Kao, a graduate student at Stanford.
“We were able to capture nuanced and nonliteral interpretations of number words using a computational model.”
When will computers understand Shakespeare?
The research shows that if listeners are trying to understand the topic and goal of communication as well as the underlying subtext—that which is not expressed explicitly—they’re better able to truly understand the utterance.
A sense of common knowledge about what’s being described or expressed is also important. Speakers and listeners assume that individuals are rational agents who use common ground and reference points to best maximize information.
As Kao puts it, “There is still a long way to go before computers can understand Shakespeare, but it is a start.”
Goodman offers this example: Imagine someone describing a new restaurant, and she says, “It took 30 minutes to get a table.” People are most likely to interpret this to mean she waited about 30 minutes.
But if she says, “It took a million years to get a table,” people will probably interpret this to mean that the wait was shorter than a million years, but that the person thinks it was much too long.
“One of the most fascinating facts about communication is that people do not always mean what they say—a crucial part of the listener’s job is to understand an utterance even when its literal meaning is false,” the researchers write.
Goodman and the others say they believe that the same ideas and techniques can extend to metaphor, irony, and many other uses of language. For example, they have a promising initial exploration of “is a” metaphors such as “your lawyer is a shark,” Goodman says.
“Beyond these cases of figurative speech, the overall mathematical framework is beginning to give a precise theory of natural language understanding that takes into account context, intention, and many subtle shades of meaning,” he says.