by Eduardo Karas
While the idea of a lie detector test has existed for a long time in the public consciouness, it was largely limited to the image of police interrogations and heartbeat monitoring. However, reseachers from Cardiff University and the Charles III University of Madrid have developed a very different kind of lie detector. With the use of machine learning and language processing, they created a model designed to identy false robbery reports, which is already being used by the Spanish National Police.
The system, called VeriPol, was trained with over 1000 reports from 2015, identifying the elements that were more likely to appear in truthful and false statements. It then produced a probability of the given report to be fabricated, based on their use of emotional language, focus on the object rather than how the robbery happened, and lack of other witnesses or doctors. To ensure the software was functioning as intended, it was subjected to comparison to two experts in identyfiying such documents from the Spanish Police. The machine was 17% more accurate in relation to the two officers.
After that, a trial study was done in conjuction with the local police of two Spanish urban areas, Murcia and Malága. The research team and specialists from the national force installed VeriPol in local departments and its force was instructed on how to use it. New reports were run through the system, which then assigned high priority to the statements that were more likely to be false and the local force would then interrogate the suspects. This method was succesful, with 83% of the cases being closed as a result.
Due to its successful pilot run, VeriPol is now being used by the Spanish National Police. While it is still a new tool, the tendency is for it to become more and more useful as officers become more used to it. As well as possibly deterring new false reports to be filled in the first place, as they will be more and more likely to be identified. In this context, it is important to note that this software is the first of its kind to be adopted on this scale, being a hallmark for both the Spanish Police and the universities involved with its research. Moreover, this also shows that lie detection for text is still a very new development, being very far from the impact of its interrogation focused cousin.