"People move more when they’re lying"

Sophie van der Zee is an assistant professor at the Erasmus School of Economics and is conducting research to develop a foolproof lie detection system, something the police could use in interrogations. In the process, though, she has identified Donald Trump’s telltale signs, and can even tell when he’s lying in a tweet.

TEXT: Suzanne Rethans

ILLUSTRATION: Claudie de Cleen

“Don't try playing amateur detective,” says academic researcher Sophie van der Zee. If there’s one thing she’s learned in the ten years that she’s been researching the most foolproof method of lie detection, it’s that you can’t tell with the naked eye. Signs like looking away and fidgeting don’t tell you anything. “We often look away when we’re thinking, whereas someone who’s lying will likely look you in the eye in the hope of convincing you they’re telling the truth.” The request that led to this research originally came from the Centre for the Protection of National Infrastructure (cpni.gov.uk), in 2010, asking: how can police interrogations be improved? How do you determine trustworthiness? Should you focus on non-verbal cues — what people do — or on what they say? Van der Zee hypothesised that non-verbal cues would be more telling, reasoning that when she herself tells a lie, she typically has a fair degree of control over what she says, but not over what her body does.

Lying for a good cause

Unfortunately, our bodies don’t respond as obviously to lying as Pinocchio’s does. “You’re actually searching for the holy grail of cues, the human version of Pinocchio’s nose,” she says. But there’s no such thing. “We might raise our chins a bit when we lie or wiggle our fingers, but nothing stands up to scrutiny once you start going through the pile of possible signs of lying.” That was the conclusion from years of studying video footage of truth-tellers and liars. All potential cues were coded and monitored: moves right hand, glances sideways, changes position. But these could just as well have been indications of other motives. It makes a difference if you’re gesticulating wildly while telling a story or lying for a good cause.

In 2011, Van der Zee learned that Ronald Poppe, a computer science researcher at the University of Twente, was hoping to generate algorithms on human behaviour from image sequences, so that movements could be measured objectively and automatically. She teamed up with him to develop a method for automatically measuring physical expressions of human behaviour and apply it to lie detection. So began a fruitful collaboration at the cutting edge of psychology and computer science. And not only in an academic sense, as they’ve since married and had a baby girl.


  • NAME: Sophie van der Zee 

    EDUCATION: Bachelor’s degree in applied economics, Erasmus School of Economics

    FUNCTIONS: 2010–2013: PhD in psychology, Lancaster University (UK)

    2013–2015: Postdoc researcher — computer science, Cambridge University (UK)

    2015–2016: Medior researcher — cybercrime, TNO

    2016–2017: Psychology lecturer, VU Amsterdam

    2018–present: assistant professor in behavioural economics at Erasmus University

All sorts of things happen under the table

They repeated her original research, but this time with subjects wearing full-body motion capture suits lined with highly sensitive sensors. The previous footage had barely shown what people were doing with their lower halves, because the subjects were seated at tables that concealed these parts. But the motion capture suits revealed that when lying, most people moved more, and moved various parts of their body, including their legs and feet. But the movements were imperceptible to the eye. Which meant motion capture suits were necessary to detect lying, but these are quite expensive. There are alternatives, such as the Xbox Kinect, but they are less accurate. Moreover, would motion capture suits still be foolproof if the wearer knew they were being monitored? Lawyers could easily instruct their clients to remain as still as possible when questioned. Thus in their follow-up research, they monitored two groups: one with subjects who were told what the suits detected and one with subjects who were told something else. Both groups were given the same task: try to beat the suit. It turned out the group that was kept in the dark moved more when lying, as previously revealed, while those who knew what was being monitored modified their behaviour to conceal their lies, but were unable to do so convincingly.

Lying is harder than telling the truth

Then, in 2013, they took the show to Cambridge, where Van der Zee was able to study the dynamics between two subjects when one of them was lying. The rationale for this was that we always lie to a second person. It turns out we mirror the person we’re talking to more when we’re lying. We don’t assume the other person believes what we’re saying, so we do things to convince them we’re telling the truth, including exhibiting agreeable behaviour. The person lying experiences emotions such as guilt and fear that they will be found out, that is unless they have psychopathic traits. Lying imposes a cognitive burden in that it is more difficult than telling the truth as you cannot draw on your memory, although that doesn’t always apply. If you have to tell a friend that her husband is cheating on her, you might find it easier to lie. And since liars want to be seen as honest, what happens when someone starts to believe their own lies? Then it’s no longer lying.

Trump knows when he’s tweeting a lie

Since 2015, Van der Zee has focused on multimodal lie detection and language analysis. Do people use language differently when they lie? A case in point: she has analysed Trump's lies on the basis of his tweets, in collaboration with fact checkers from the Washington Post. That he states factual inaccuracies is clear. But does he know that he is lying or is he just poorly informed? Van der Zee’s linguistic analysis revealed that he uses very different words in factually accurate tweets when compared with those that appear in his factually inaccurate ones. Which means he usually knows when he is lying. Implementing these findings in real-life applications is still difficult. A multimodal approach works best: the combination of non-verbal, verbal and physiological cues. But several authorities have expressed interest, including not just the police, but also border services and the insurance industry.

“When we’re lying, we don’t assume the other person believes what we’re saying, so we do things to convince them we’re telling the truth, including exhibiting agreeable behaviour.”