Scientists have created VeriPol, a lie detector for written texts. The computer tool can tell whether somebody has filed a false police statement. It detects the lie just by analyzing a written text.
According to the University of Cardiff, VeriPol is over 80% accurate. It can successfully identify, for example, false robbery reports more than eighty percent of the time.
The tool uses a combination of advanced machine learning techniques and automatic text analysis.
Police officers in Spain are using VeriPol. They are using it to support their work and also to suggest where further investigations are necessary.
VeriPol specifically for robbery reports
The lie detector is specific to robbery reports. It can recognize patterns that commonly exist in false claims, such as the kinds of items the thief took. It can also recognize patterns in descriptions of the perpetrator and other finer details of incidents.
The research team, from Universities in Cardiff and Madrid, wrote about their work in the journal Knowledge-Based Systems (citation below).
They believe that the tool could save the police valuable time and effort by complementing existing investigative techniques. It could also deter lying. In other words, if people know what VeriPol can do, they are more likely to tell the truth.
In most countries, filing false police statements can lead to either heavy fines or jail terms.
Lies undermine the outcomes of criminal investigations and contaminate police databases. They also waste precious public resources that could otherwise be used to pursue real crimes.
Despite the deterrents, false reports are extremely common, especially in cases of low-level crimes, such as robbery.
Natural language processing
The creators of VeryPol based it partly on ‘natural language processing.’ Natural language processing is an AI branch that helps computers understand, interpret and manipulate language; specifically human language.
AI stands for Artificial Intelligence, i.e., software technology that makes computers think like humans. It also makes them behave like humans.
The computer tool uses algorithms to identify and quantify several features in the text. It identifies and quantifies, for example, acronyms, adjectives, nouns, verbs, numbers, figures, and punctuation marks.
False reports were fed through VeriPol so that it could code each one and start ‘learning’ the specific patterns.
Researchers carried out an initial study on over 1,000 police reports from the Spanish National Police. VeriPol was ‘extremely effective’ in telling which reports were true and which were false. It had a success rate of over eighty percent.
False robbery reports – characteristics
VeriPol was able to identify several characteristics that are common in false reports. A short statement that focuses more on the stolen items than the incident, for example, is a common characteristic of false reports.
A lack of witnesses or other hard evidence, limited details of the attacker, and a lack of precise details about the incident are also common traits.
Co-author, Dr Jose Camacho-Collados, said:
“As an example, our model began to identify false statements where it was reported that incidents happened from behind or where the aggressors were wearing helmets.”
Dr. Camacho-Collados is a Research Associate at Cardiff University’s School of Computer Science and Informatics.
VeriPol vs. humans
In June 2017, VeriPol detected 25 cases of false robbery in Murcia, Spain, in just one week. From 2008 to 2016, the police in Murcia detected an average of 3.33 false reports in the month of June. The police in Malaga, during the same month over the same period, detected only (average) 12.14 false reports.
Dr Camacho-Collados said:
“Our study has given us a fascinating insight into how people lie to the police, and a tool that can be used to deter people from doing so in the future,” continued Dr Camacho-Collados.”
“Police officers across Spain are now using VeriPol and integrating it into their working practices. Ultimately we hope that by showing that automatic detection is possible it will deter people from lying to the police in the first instance.”
“Applying automatic text-based detection of deceptive language to police reports: Extracting behavioral patterns from a multi-step classification model to understand how we lie to the police,” Lara Quijano-Sánchez, Federico Liberatore, José Camacho-Collados, and Miguel Camacho-Collados. Knowledge-Based Systems, Volume 149, Pages 155-168. DOI: https://doi.org/10.1016/j.knosys.2018.03.010.