Detecting Children’s Deception Through Facial Expressions

What is it about?

This study examined whether children’s facial expressions could provide clues about when they are being deceptive. The researchers studied 158 children between 4 and 9 years old. Some children experienced two toys appearing to break while playing with an adult and were then asked by the adult to keep the broken toys secret. Children who later concealed the broken toys during an interview were classified as liars. A comparison group of children played with the toys but did not experience any toys breaking, so they had nothing to conceal. The researchers used computer software to automatically analyze children’s facial expressions and machine learning to determine whether those expressions could distinguish the two groups. Facial expressions displayed while children were hearing the interviewer’s question distinguished liars from nonliars with 73% overall accuracy. In particular, children who concealed the broken toys showed more facial expressions associated with surprise and fear. These differences between liars and nonliars did not depend on children’s age, gender, or history of maltreatment. However, facial expressions were not able to reliably distinguish the groups during the period after the question had been asked, when children were preparing their response.

Featured Image

Photo by Lora Moore-Kakaletris on Unsplash

Photo by Lora Moore-Kakaletris on Unsplash

Why is it important?

Adults, including professionals with experience interviewing children, often have difficulty determining when children are being deceptive. This study suggests that automated facial-expression technology may be able to detect subtle differences in children’s expressions that people would be unlikely to notice themselves. It also provides researchers with a noninvasive way to examine the emotions children may experience while concealing information. However, the findings are an early step rather than a practical lie-detection tool. Although the computer model achieved 73% overall accuracy, it was much better at identifying liars than nonliars, correctly classifying 92% of liars but only 35% of nonliars. In addition, the children in the two groups had different experiences: children who lied had experienced the toys breaking, whereas the comparison group had not. This means that differences in fear and surprise cannot yet be attributed specifically to lying. The authors emphasize that the findings need to be replicated across different situations before automated facial expressions could be used to identify deception in real-world settings.