After analyzing 50,389 verbal exchanges from 204 Supreme Court cases and 240,000 conversations among Wikipedia editors, four Cornell computer scientists recently discovered that people unconsciously mimic the linguistic style of those whom they view as socially superior.
“We find that power –– and in particular, status –– differences between people are subtly revealed by their conversational behavior,” said Cristian Danescu-Niculescu-Mizil grad, the lead author on the study. “When a person gains higher status –– through a promotion, for instance –– people will react to this change in status by increasing how much they echo the style of the person that undergoes status change.”
The co-authors of the research are Prof. Jon Kleinberg ’93, computer science; Prof. Lillian Lee ’93, computer science; and Yahoo! researcher Bo Pang Ph.D. ’06.
The Cornell team analyzed the language used on Wikipedia talk pages, in which writers and editors discuss their articles, as well as the language used in Supreme Court cases.
“In the Supreme Court transcripts, there’s a set of language effects unfolding below the overt arguments, giving us extra clues as to which side each Justice is favoring,” Kleinberg said.
Danescu-Niculescu-Mizil said that the researchers tried to compare two different areas of conversation.
“We wanted to understand whether our findings apply in very different domains and settings,” Danescu-Niculescu-Mizil said. “Wikipedia discussions take place online, in textual format, while the oral arguments in the Supreme Court are off-line and the interaction takes place through speech.”
Danescu-Niculescu-Mizil explained that the study analyzed how people construct what they want to say, as opposed to the particular diction they employ.
“We look at features of language that have to do with how things are said, as opposed to what is said,” he said. “For example, we look at the usage of articles, prepositions, conjunctions and quantifiers.”
Danescu-Niculescu-Mizil presented the study at the World Wide Web Conference — a yearly international academic conference on the direction of the Internet — which took place on April 16 to April 20 in Lyon, France.
“Presenting at the World Wide Web conference is a fun and rewarding experience,” he said. “I have been lucky to present there four years in a row. It is always an honor to represent Cornell at such an important venue.”
Danescu-Niculescu-Mizil said that his research could potentially lead to an understanding of who should take charge in situations in which there is not a clear leader.
“From a practical perspective, future work based on these findings could lead to systems that can automatically infer power relations, and perhaps other types of relations, between users of social media platforms,” he said. “These findings could also lead to a way to discover [who is in charge], when nobody is in charge.”
Similarly, Kleinberg said that although many people do not realize that they change their syntax, their analysis provides insight into identifying authoritative figures.
“It’s striking how this effect is operating largely outside our conscious perception, but when you ... start measuring it, it can tell you things about who holds power and who’s acting differentially,” he said.
Danescu-Niculescu-Mizil also said that he was surprised how some features of language, which many assume to be negligible, can actually provide sociological data.
“Something I find fascinating about these results is that they reveal how apparently innocuous features of language, such as the usage of prepositions or articles, can tell a lot about one’s perception of self and of the others … much more than one would have expected,” he said.