You might expect a liar to struggle with eye contact, when lying through his teeth. But a study by University of Michigan researchers suggests that liars may look their questioners in the eye more often than people telling the truth. People who are lying also wave both hands around far more than those who are being honest.
The researchers studied 118 video clips to establish the language and gestures used by people being dishonest, including testimonies from the Innocence Project, a non-profit that the handles legal cases of innocent people falsely imprisoned. Around half the clips came from trial footage, while the other half focused on YouTube videos asking interviewees their opinion on films. Researchers established who was lying by the trial verdict (with the exception of the Innocence Project clips), and when interviewees were asked their opinions on films that didn’t exist.
Those who were lying were found more likely to have animated hand movements, make strong eye contact, nod their heads, and scowl.
When researchers transcribed the audio, they also found that liars were more likely to say “um” and to use pronouns that distanced themselves from the action, such as “he” or “she” rather than “I” or “we.”
The same gestures were also used by some people telling the truth, but to a lesser extent: 25% of truthful people gestured with both hands, compared to 40% of liars. Sixty percent of truthful people looked directly at the questioner, compared to 70% of liars.
The real world nature of the footage was important to the study said Rada Mihalcea, professor of computer science and engineering who co-led the project, in a statement. She added:
“In laboratory experiments, it’s difficult to create a setting that motivates people to truly lie. The stakes are not high enough. We can offer a reward if people can lie well—pay them to convince another person that something false is true. But in the real world there is true motivation to deceive.”
When the researchers compiled all the data into machine-learning software, the computer was 75% accurate at identifying who was lying. Humans only guessed right 50% of the time.
The paper was presented at the International Conference on Multimodal Interaction, which has a peer-reviewed selection process, in November and is published in the 2015 conference proceedings.
The researchers believe their work could be useful for security agents and juries, but of course it’s far from foolproof. It’s also impossible to know for sure whether the courtroom verdicts are right, and if the “liars” being examined are definitely being dishonest. This is a problem for all lie detection; Polygraph tests, which measure stress in response to questions, have similarly variable results. Though they have roughly 85% accuracy when testing guilty people, some polygraph tests have just 56% accuracy for innocent people.
The truth is, we all behave slightly differently when we’re dishonest, and there’s no definitive telltale sign that unmasks all liars. But the University of Michigan research can help build a picture of dishonesty. And, as Mihalcea points out, the software is certainly better than humans at picking up on falsehoods.