New Software Detects Lies by Analysing Words and Gestures
Friday, December 11, 2015
Researchers from the University of Michigan are developing a lie detecting software that analyses the words and gestures of the speaker, unlike a polygraph test wherein the machine is supposed to touch the speaker in order to function.
Michigan scientists, for this purpose, are studying high profile court cases wherein they found out that those who lied moved their hands more and looked into questioners’ eye for a longer duration than those who spoke the truth. It was also observed that those who lied spoke with more vocal fill such as ‘umm’.
The machine had an accuracy rate of about 75 percent in detecting who was being deceptive (defined in the final outcome of a case) when compared to humans’ score of just 50 percent. Scientists are using machine learning techniques to train the software on a set of 120 video clips from the media coverage of actual trials.