Software Being Developed to Monitor Opinions of U.S.
By ERIC LIPTON
Published: October 4, 2006
WASHINGTON, Oct. 3 — A consortium of major universities, using Homeland
Security Department money, is developing software that would let the
government monitor negative opinions of the United States or its leaders in
newspapers and other publications overseas.
Such a “sentiment analysis” is intended to identify potential threats to the
nation, security officials said.
Researchers at institutions including Cornell, the University of Pittsburgh
and the University of Utah intend to test the system on hundreds of articles
published in 2001 and 2002 on topics like President Bush’s use of the term
“axis of evil,” the handling of detainees at Guantánamo Bay, the debate over
global warming and the coup attempt against President Hugo Chávez of
A $2.4 million grant will finance the research over three years.
American officials have long relied on newspapers and other news sources to
track events and opinions here and abroad, a goal that has included the
routine translation of articles from many foreign publications and news
The new software would allow much more rapid and comprehensive monitoring of
the global news media, as the Homeland Security Department and, perhaps,
intelligence agencies look “to identify common patterns from numerous
sources of information which might be indicative of potential threats to the
nation,” a statement by the department said.
It could take several years for such a monitoring system to be in place,
said Joe Kielman, coordinator of the research effort. The monitoring would
not extend to United States news, Mr. Kielman said.
“We want to understand the rhetoric that is being published and how intense
it is, such as the difference between dislike and excoriate,” he said.
Even the basic research has raised concern among journalism advocates and
privacy groups, as well as representatives of the foreign news media.
“It is just creepy and Orwellian,” said Lucy Dalglish, a lawyer and former
editor who is executive director of the Reporters Committee for Freedom of
Andrei Sitov, Washington bureau chief of the Itar-Tass news agency of
Russia, said he hoped that the objective did not go beyond simply
identifying threats to efforts to stifle criticism about an American
president or administration.
“This is what makes your country great, the open society where people can
criticize their own government,” Mr. Sitov said.
The researchers, using an grant provided by a research group once affiliated
with the Central Intelligence Agency, have complied a database of hundreds
of articles that it is being used to train a computer to recognize, rank and
The software would need to be able to distinguish between statements like
“this spaghetti is good” and “this spaghetti is not very good — it’s
excellent,” said Claire T. Cardie, a professor of computer science at
Professor Cardie ranked the second statement as a more intense positive
opinion than the first.
The articles in the database include work from many American newspapers and
news wire services, including The Miami Herald and The New York Times, as
well as foreign sources like Agence France-Presse and The Dawn, a newspaper
One article discusses how a rabid fox bit a grazing cow in Romania, hardly a
threat to the United States. Another item, an editorial in response to Mr.
Bush’s use in 2002 of “axis of evil” to describe Iraq, Iran and North Korea,
said: “The U.S. is the first nation to have developed nuclear weapons.
Moreover, the U.S. is the first and only nation ever to deploy such
The approach, called natural language processing, has been under development
for decades. It is widely used to summarize basic facts in a text or to
create abridged versions of articles.
But interpreting and rating expressions of opinion, without making too many
errors, has been much more challenging, said Professor Cardie and Janyce M.
Wiebe, an associate professor of computer science at the University of
Pittsburgh. Their system would include a confidence rating for each
“opinion” that it evaluates and would allow an official to refer quickly to
the actual text that the computer indicates contains an intense
Ultimately, the government could in a semiautomated way track a statement by
specific individuals abroad or track reports by particular foreign news
outlets or journalists, rating comments about American policies or
Marc Rotenberg, executive director of the Electronic Privacy Information
Center in Washington, said the effort recalled the aborted 2002 push by a
Defense Department agency to develop a tracking system called Total
Information Awareness that was intended to detect terrorists by analyzing
troves of information.
“That is really chilling,” Mr. Rotenberg said. “And it seems far afield from
the mission of homeland security.”
Federal law prohibits the Homeland Security Department or other intelligence
agencies from building such a database on American citizens, and no effort
would be made to do that, a spokesman for the department, Christopher Kelly,
said. But there would be no such restrictions on using foreign news media,
Mr. Kelly said.
Mr. Kielman, the project coordinator, said questions on using the software
were premature because the department was just now financing the basic
research necessary to set up an operating system.
Professors Cardie and Wiebe said they understood that there were legitimate
questions about the ultimate use of their software.
“There has to be guidelines and restrictions on the use of this kind of
technology by the government,” Professor Wiebe said. “But it doesn’t mean it
is not useful. It can just as easily help the government understand what is
going on in places around the world.”