Researcher sees patterns in North Korean news reports before military actions
By MADELEINE O'NEILL | Erie Times-News, Pa. | Published: January 1, 2018
ERIE, Pa. (Tribune News Service) — A Mercyhurst University professor says he developed a computer model that identified patterns in North Korean news reports published before military actions, such as nuclear tests or missile launches.
The program was based on analysis of the language used in the news reports, said the professor, Michael Lamm, who goes by Michael Lammbrau professionally.
"If you're about to get in a fight with somebody, the language changes," he said.
His research on the subject has caught the attention of national and international media outlets, not long after North Korean leader Kim Jong Un and President Donald Trump this fall traded insults including "dotard" and "rocket man."
A Dec. 14 report in the Wall Street Journal featured Lammbrau's work, and the Daily Express, a British news outlet, published a similar article the same day. Lammbrau said news outlets in South Korea, China and Vietnam have also reported on his research.
Lammbrau said the work highlights the importance of language and how it can affect international relations.
"What the research shows is that rhetoric can and does predict conflict," Lammbrau said. He hopes to "bring a little bit of responsibility to news outlets and leaders around the world for their language."
Lammbrau developed the computer program while pursuing his Ph.D. at the University of North Korean Studies, in Seoul. The Mercyhurst graduate had previously spent more than two years in South Korea serving as a Korean linguist in the U.S. Army.
"We were interested in trying to predict or gain insight into when North Korea was going to do a nuclear test or launch a missile," he said.
His research harnessed machine learning to build a model that could identify key language in North Korean news reports that might unknowingly give away an upcoming action, like a missile launch.
"The idea is to identify predictive features ... in the text that occurs prior to an event," Lammbrau said.
In tests, the computer program accurately predicted nuclear tests by North Korea 73.2 percent of the time, according to the research. The model also successfully identified "noise," or North Korean rhetoric that did not result in conflict, about 96 percent of the time.
Lammbrau is currently working with Mercyhurst students to develop a model that could use online media reports to predict when the probability of conflict is heightened in countries around the world.
"The goal of the research is to reduce the possibility of conflict," he said.
©2018 the Erie Times-News (Erie, Pa.)
Visit the Erie Times-News (Erie, Pa.) at www.GoErie.com
Distributed by Tribune Content Agency, LLC.