Queen’s own spin doctor

Computer model detects deception in political speeches

Queen’s Computing Science professor David Skillicorn says campaign advertisements are harder to analyze than speeches because they’re more staged.
Queen’s Computing Science professor David Skillicorn says campaign advertisements are harder to analyze than speeches because they’re more staged.

In the throes of two major elections, it can be tough to detect which politicians are genuine and which are spinning tall tales. But according to a model developed by a Queen’s professor, sniffing out spin is as easy as crunching a few numbers.

David Skillicorn, a professor in the School of Computing, analyzes political speeches to determine how truthful a candidate’s words are. Skillicorn has examined speeches by American Republican presidential candidate John McCain and Democratic candidate Barack Obama. In comparing speeches from both parties’ conventions, he found Obama’s speech had a higher level of spin than did McCain’s.

Skillicorn said people lie on a spectrum. There are outright, deceptive lies, but there are also more socially acceptable forms of lying. Politicians live somewhere in the middle of that spectrum, he said.

“Politicians are always trying to present himself or herself as the most attractive person they can possibly be, and that’s a form of deceptiveness,” he said.

But that form of political deception is a conscious choice politicians make, Skillicorn said. His research focuses on what’s primarily unconscious and comes across in large, publicized events such as political speeches, as well as in everyday situations such as job interviews.

“What you’re going to get is a variation on what that person is actually like that makes them look just a little bit nicer and better and stronger than they really are,” he said.

Although he works with computing science, psychology is where much of Skillicorn’s model is based. He said his work comes from the research of James Pennebaker, a psychologist at the University of Texas at Austin.

The difference between his model and Pennebaker’s research is the new focus on correlation information, Skillicorn said.

Skillicorn’s model is based on speech patterns among four classes of words. He said deceptive speech has fewer first-person singular pronouns and exclusive words, and an increase in negative emotion words and action words.

When somebody’s being deceptive, they use fewer singular pronouns such as ‘I,’ he said.

“Psychologically, they’re kind of stepping back a little bit from what they’re saying—distancing themselves from it—and that shows up from this reduction in pronoun use.” Exclusive words introduce some variation to a story, Skillicorn said.

“A part of your brainpower is being spent to create this image that isn’t quite the unvarnished truth, and that means you have to keep things a little simpler than you otherwise would,” he said. “It’s really hard to keep adding complexities to whatever it is that you’re talking about.”

He said people use negative emotion words because it’s socially frowned upon to be deceptive.

“That negative feeling that they have about themselves leaks out a bit into the words they use,” he said.

He said the fourth change—an increase in action words—is the most puzzling.

“It has to do with keeping the story moving, trying to ... push the listener over any bumps there might be in something that doesn’t quite fit or because you’ve constructed it on the fly, it doesn’t look quite as perfectly smooth as you’d like it to be,” he said, adding that the first three categories seem more plausible than the fourth on psychological grounds.

Although his program analyzes just over 80 words sending different signals to the listener, some words are more important than others in different contexts.

“The [words] that vary the most in the set of things you’re looking at will tend to tell you more information,” he said.

Skillicorn said examining the frequency of words used in speech is no easy task for the human ear.

“We don’t have that much control over the way we use language, not consciously,” he said.

He said other researchers examine speech patterns in terms of factors such as facial expression and voice texture to determine whether someone’s being deceptive, but much of that can be controlled consciously. Orators often go through training to learn how to control these other factors, but because spin is primarily unconscious, it’s difficult to modify.

“You could … go through your speech before you gave it and try and change things so that it would score a little bit more the way that you wanted it to,” he said. “But the trouble is what seems to happen is when you actually give the speech, you put all that stuff back again.”

“The only way you can change it is to change the way you frame the whole situation you’re in,” he said, citing Hillary Clinton—whose speech patterns changed when it became clear Obama was going to win the Democratic nomination—as an example.

Skillicorn said what politicians choose to talk about has a strong influence on how they’ll score. When a speech has a personal bent, it’s different than a speech on a broad issue and is less susceptible to spin.

“If you want to say, ‘I’m going to put giraffes on Mars because I’m an expert on space travel,’ then you’re almost forced to associate who you are to that part of that story,” he said.

Skillicorn said the Canadian election has been harder to analyze because the candidates haven’t posted enough speeches online, but Prime Minister Stephen Harper’s eight posted speeches have varied in their level of spin.

“You don’t have to look at them very hard to see that they’re very different from what’s going on in the U.S. situation,” he said. “Harper’s speeches are almost like press releases.”

Campaign advertisements are also harder to analyze, Skillicorn said, because they’re more staged than speeches.

“The more off the cuff it is, the more revealing it is,” he said.

Acknowledging many campaign speeches are speechwriters’ collaborations, Skillicorn said candidates still change their deliveries depending on how they’re feeling at the time.

“When a candidate’s reading from a teleprompter, they change on the fly some of these little words which are exactly the ones the model’s picking up,” he said. “Even the speechwriters tend not to notice that that’s happened.”

He said it’s more difficult to measure how deception comes into play for public figures in question-and-answer situations.

“If somebody says to you, ‘What are you going to do?’ you’re almost forced to say, ‘I am going to do this and that,’ … which means that now you’re sending a signal which looks like a lower level of spin.”

Skillicorn said his model can’t be applied to speech in different contexts—for example, it can’t compare a politician and a car salesman.

“The problem is that all these things rely on changes in the relative frequency of words, and so you have to kind of know what the norm is,” he said.

In business writing, for example, it’s much more common to use the pronoun ‘I’ than it might be in other situations.

Skillicorn said his model deals with the use of the word ‘I,’ not ‘we.’ He said politicians are trained to use ‘we,’ but psychologists have found that when men use the word ‘we’ it’s not inclusive at all. He said Obama uses ‘we’ much more often than McCain.

“That’s very much the velvet glove around the iron fist, and it’s a euphemism for ‘you,’ a command. So Obama’s ‘we’s’ are a dangerous thing for him to use, in a way.

“‘We’ can be collective responsibility, but you should always stop to think about it. Could you put ‘you’ here? If you can, then it’s a fake.”

Skillicorn said his model could be applied to everyday conversation, but at this point researchers don’t know how.

He has received requests from as far away as the United Kingdom, Spain and Portugal to apply his program to politicians. But he said the model doesn’t easily translate to other languages.

“In French you don’t know what to do with the pronoun ‘on,’ because sometimes … it’s equivalent to ‘je’ and sometimes it’s not, and it’s really hard to tell the difference,” he said. Skillicorn also said it would be difficult to apply his model to historical speeches because of changes in the use of language and a lack of political context.

“The problem with all of this is getting some estimate of the ground truth,” he said. “That gets harder as you go back in time because you just don’t know the political context nearly as well.”

See more results from Skillicorn’s research at skillicorn.wordpress.com.

Spin Doctor's Scale

Based on speeches made at the Democratic and Republican conventions, Skillicorn measured spin for a number of American politicians. Here are the results:

Most deceptive to least deceptive

4.36 - Mitt Romney
2.43 - Hillary Clinton
0.99 - Bill Clinton
0.40 - George Bush
0.31 - Barack Obama
-0.62 - Sarah Palin
-1.24 - Michelle Obama
-1.35 - Joe Biden
-7.38 - John McCain

In so many words

Computing Science Professor David Skillicorn analyzes speeches made by politicians such as American presidential candidates Democrat Barack Obama and Republican John McCain. Here are some samples of “typical” speeches made by Obama and McCain:

Barack Obama (Feb. 13, 2008)
“That our prosperity can and must be the tide that lifts every boat; that we rise or fall as one nation; that our economy is strongest when our
middle-class grows and opportunity is spread as widely as possible. And when it’s not—when opportunity is uneven or unequal—it is our responsibility to restore balance, and fairness, and keep that promise alive for the
next generation.”

John McCain (May 27, 2008)
“I believe we must also address nuclear testing. As president I will pledge to continue America’s current moratorium on testing, but also begin a dialogue with our allies, and with the U.S. Senate, to identify ways we can move forward to limit testing in a verifiable manner that does not undermine the security or viability of our nuclear deterrent.”

All final editorial decisions are made by the Editor(s)-in-Chief and/or the Managing Editor. Authors should not be contacted, targeted, or harassed under any circumstances. If you have any grievances with this article, please direct your comments to journal_editors@ams.queensu.ca.

When commenting, be considerate and respectful of writers and fellow commenters. Try to stay on topic. Spam and comments that are hateful or discriminatory will be deleted. Our full commenting policy can be read here.