Lie detection is an assessment of a verbal statement with the goal to reveal a possible intentional deceit. Lie detection may refer to a cognitive process of detecting deception by evaluating message content as well as non-verbal cues. It also may refer to questioning techniques used along with technology that record physiological functions to ascertain truth and falsehood in response. The latter is commonly used by law enforcement in the United States, but rarely in other countries and has historically been an inexact science. There are a wide variety of technologies available for this purpose. The most common and long used measure is the polygraph, which the U.S. National Academy of Sciences states, in populations untrained in countermeasures, can discriminate lying from truth telling at rates above chance, though below perfection. They added that the results apply only to specific events and not to screening, where it is assumed that the polygraph works less well.
- 1 General accuracy and limitations of assessment
- 2 History
- 3 General questioning and testing techniques
- 4 The control question test and the guilty knowledge test
- 5 Polygraph
- 6 Cognitive polygraph
- 7 ERP
- 8 EEG
- 9 Eye-tracking
- 10 Voice stress analysis
- 11 FMRI
- 12 fNIRS
- 13 Brain observations
- 14 Nonverbal behavior
- 15 Drugs
- 16 See also
- 17 References
General accuracy and limitations of assessment
The cumulative research evidence suggests that machines do detect deception better than chance, but with significant error rates and that strategies used to “beat” polygraph examinations, so-called countermeasures, may be effective. Despite unreliability, results are admissible in court in some countries such as Japan. Lie detector results are very rarely admitted in evidence in the US courts.
In 1983 the U.S. Congress Office of Technology Assessment published a review of the technology and found:
“…there is at present only limited scientific evidence for establishing the validity of polygraph testing. Even where the evidence seems to indicate that polygraph testing detects deceptive subjects better than chance, significant error rates are possible, and examiner and examinee differences and the use of countermeasures may further affect validity.”
In the 2007 peer-reviewed academic article “Charlatanry in forensic speech science”, the authors reviewed 50 years of lie detector research and came to the conclusion that there is no scientific evidence supporting that voice analysis lie detectors actually work. Lie detector manufacturer Nemesysco threatened to sue the academic publisher for libel resulting in removal of the article from online databases. In a letter to the publisher, Nemesysco’s lawyers wrote that the authors of the article could be sued for defamation if they wrote on the subject again. The full text of the article is available online.
Extraneous noise can come from embarrassment or anxiety and not be specific to lying. When subjects are aware of the assessment their resulting emotional response, especially anxiety, can impact the data. Additionally, psychological disorders can cause problems with data as certain disorders can lead a person to make a statement they believe to be truth but is actually a fabrication. As well as with all testing, the examiner can cause biases within the test with their interaction with the subject and interpretation of the data.
The study of physiological methods for deception tests measuring emotional disturbances began in the early 1900s. Vittorio Benussi was the first to work on practical deception tests based on physiological changes. He detected changes in inspiration-expiration ratio—findings confirmed by N.E. Burtt. Burtt conducted studies that emphasized the changes in quantitative systolic blood-pressure. William Moulton Marston studied blood-pressure and noted increase in systolic blood pressure of 10 mm Hg or over indicated guilt through using the tycos sphygmomanometer, with which he reported 90-100% accuracy. His studies used students and actual court cases. Then in 1913 W.M. Marston determined systolic blood-pressure by oscillatory methods and his findings cite definite changes in blood pressure during the deception of criminal suspects. In 1921, John Augustus Larson criticized Marston’s intermittent blood pressure method because emotional changes were so brief they could be lost. To adjust for this he modified the Erlanger sphygmograph to give a continuous blood pressure and pulse curve and used it to study 4,000 criminals.
A meta-analysis study from DePaulo and Morris found an “association between lying and increased pupil size, an indicator of tension and concentration.” Additionally, those lying are perceived to appear more nervous than those telling the truth (which may be because the voices of those lying are higher pitched), while they also do not appear to be more fidgety, blink more, or have a less-relaxed posture but “are more likely than truth-tellers to press their lips together.” However, highly motivated liars (those with higher stakes) “seem unusually still and make notably less eye contact with listeners.”
Paul Ekman has used the Facial Action Coding System (FACS) and “when combined with voice and speech measures, [it] reaches detection accuracy rates of up to 90 percent.” However, there is currently no evidence to support such a claim. It is currently being automated for use in law enforcement and is still being improved to increase accuracy. His studies use micro-expressions, which last less than one-fifth of a second, and “may leak emotions someone wants to conceal, such as anger or guilt.” However, “signs of emotion aren’t necessarily signs of guilt. An innocent person may be apprehensive and appear guilty” Ekman reminds us. With regard to his studies, lies about emotions at the moment have the biggest payoff from face and voice cues while lies about beliefs and actions, such as crimes use cues from gestures and words are added. Ekman and his associates have validated many signs of deception, but do not publish all of them as not to educate criminals
Depaulo and her graduate student Morris have been studying the verbal and written output of liars to find distinctive patterns. They have found that “liars take longer to start answering questions than truth-tellers–but when they have time to plan, liars actually start their answers more quickly than truth-tellers. And they talk less.” When considering the perception of others, “liars seem more negative–more nervous and complaining, and less cooperative–than truth-tellers” and they additionally seem to withhold more information. Individuals lying sound “more discrepant and ambivalent, the structure of their stories is less logical, and their stories sound less plausible.” Additionally, it has been observed that they are more likely than those telling the truth to repeat words and phrases, but they also use fewer hand movements to aid in the description of their actions.
James Pennebaker uses the method of Linguistic Inquiry and Word Count (LIWC), published by Lawrence Erlbaum, to conduct an analysis of written content. He claims it has accuracy in predicting lying. Pennebaker cites his method as “significantly more effective than human judges in correctly identifying deceptive or truthful writing samples”; there is a 67% accuracy rate with his method, while trained people have 52% accuracy. There were five experimental procedures used in this study. Study 1-3 asked participants to speak, hand write or type a true or false statement about abortion. The participants were randomly assigned to tell a true or false statement. Study 4 focused on feelings about friends and study 5 had the students involved in a mock crime and asked to lie. Human judges were asked to rate the truthfulness of the 400 communications dealing with abortion. The judges read or watched the statement and gave it a yes or no answer about if this statement was false or not. LIWC correctly classified 67% of the abortion communications and the judges correctly classified 52%. His studies have identified that deception carries three primary written markers. The first is fewer first-person pronouns. Those lying “avoid statements of ownership, distance themselves from their stories and avoid taking responsibility for their behavior” while also using more negative emotion words such as “hate, worthless and sad.” Second, they use “few exclusionary words such as except, but or nor” when “distinguish[ing] what they did from what they did not do.”
More recently evidence has been provided by the work of CA Morgan III and GA Hazlett that a computer analysis of cognitive interview derived speech content (i.e. response length and unique word count) provides a method for detecting deception that is both demonstrably better than professional judgments of professionals and useful at distinguishing between genuine and false adult claims of exposure to highly stressful, potentially traumatic events. This method shows particular promise as it is non confrontational as well as scientifically and cross culturally valid.
General questioning and testing techniques
There are typically three types of questions used in polygraph testing or voice stress analysis testing:
Irrelevant questions establish a baseline to compare other answers by asking simple questions with clear true and false answers.
Comparison questions have an indirect relationship to the event or circumstance, and they are designed to encourage the subject to lie.
Relevant questions are compared against comparison questions (which should represent false answers) and irrelevant questions (which should represent true answers). They are about whatever is particularly in question
The control question test and the guilty knowledge test
The control question test (CQT) uses control questions, with known answers, to serve as a physiological baseline in order to compare them with questions relevant to a particular incident. The control question should have a greater physiological response if truth was told and a lesser physiological response for lying. The guilty knowledge test (GKT) is a multiple-choice format in which answer choices or one correct answer and additional incorrect answers are read and the physiological response is recorded. The controls are the incorrect alternative answers. The greater physiological response should be to the correct answer. Its point is to determine if the subject has knowledge about a particular event.
Both are considered to be biased against those that are innocent, because the guilty who fear the consequences of being found out can be more motivated to cheat on the test. Various techniques (which can be found online) can teach individuals how to change the results of the tests, including curling the toes, and biting the tongue. Mental arithmetic was found to be ineffective by at least one study, especially in students counting backward by seven. A study has found that in the guilty knowledge test subjects can focusing on the alternative answers and make themselves look innocent.
Main article: Polygraph
Lie detection commonly involves the polygraph; however it is not considered totally reliable. It detects autonomic reactions. These changes in body functions are not easily controlled by the conscious mind and include bodily reactions like skin conductivity and heart rate. They also may consider respiration rate, blood pressure, capillary dilation, and muscular movement. While taking a polygraph test the subject wears a blood pressure device to measure blood pressure fluctuations. Respiration is measured by wearing pneumographs around the chest, and finally electrodes are placed on the subject’s fingers to measure skin conductivity. To determine truth it is assumed the subject will show more signs of fear when answering the control questions compared with the relevant questions. If a person is showing a deception there will be changes in the autonomic arousal responses to the relevant questions. Results are considered inconclusive if there is no fluctuation in any of the questions.
These measures are supposed to indicate a short-term stress response which can be from lying or significance to the subject. The problem becomes that they are also associated with mental effort and emotional state, so they can be influenced by fear, anger, and surprise for example. This technique may also be used with CQT and GKT. There are many issues with polygraph tests because many people have found ways to try and cheat the system. Some people have been known to take sedatives to reduce anxiety; using antiperspirant to prevent sweating; and positioning pins or biting parts of the mouth after each question to demonstrate a constant physiological response.
Activities of the body not easily controlled by the conscious mind are compared under different circumstances. Usually this involves asking the subject control questions where the answers are known to the examiner and comparing them to questions where the answers are not known. Critics claim that “lie detection” by use of polygraphy has no scientific validity because it is not a scientific procedure. Government agencies, such as the Department of Defense, Homeland Security, Customs and Border Protection, and even the Department of Energy currently use polygraphs. They are regularly used by these agencies to screen employees. The problem with evaluating the effectiveness of polygraphs through field studies is that the use of confessions overestimates accuracy. Someone who has failed the test is more likely to confess than someone who has passed, contributing to polygraph examiners not learning about mistakes they have made and thus improving.
Recent developments that permit non-invasive monitoring using functional transcranial Doppler (fTCD) technique showed that successful problem-solving employs a discrete knowledge strategy (DKS) that selects neural pathways represented in one hemisphere, while unsuccessful outcome implicates a non-discrete knowledge strategy (nDKS). A polygraphic test could be viewed as a working memory task. This suggests that the DKS model may have a correlate in mnemonic operations. In other words, the DKS model may have a discrete knowledge base (DKB) of essential components needed for task resolution, while for nDKS, DKB is absent and, hence, a “global” or bi-hemispheric search occurs. Based on the latter premise, a ‘lie detector’ system was designed as described in United State Patent No. 6,390,979. A pattern of blood-flow-velocity changes is obtained in response to questions that include correct and incorrect answers. The wrong answer will elicit bi-hemispheric activation, from correct answer that activates unilateral response. Cognitive polygraphy based on this system is devoid of any subjective control of mental processes and, hence, high reliability and specificity; however, this is yet to be tested in forensic practice. See also cognitive biometrics.
Main article: Event-related potentials
Event-related potentials assess recognition, and therefore may or may not be effective in assessing deception. In ERP studies P3 amplitude waves are assessed, with these waves being large when an item is recognized. However, P100 amplitudes have been observed to have significant correlation to trustworthiness ratings, which the importance of will be discussed in the EEG section. This, along with other studies leads some to purport that because ERP studies rely on quick perceptual processes they “are integral to the detection of deception.”
Main article: Electroencephalography
Electroencephalography, or EEG, measures brain activity through electrodes attached to the scalp of a subject. The object is to identify the recognition of meaningful data through this activity. Images or objects are shown to the subject while questioning techniques are implemented to determine recognition. This can include crime scene images, for example.
Perceived trustworthiness is interpreted by the individual from looking at a face, and this decreases when someone is lying. Such observations are “too subtle to be explicitly processed by observers, but does affect implicit cognitive and affective processes.” These results, in a study by Heussen, Binkofski, and Jolij, were obtained through a study with an N400 paradigm including two conditions within the experiment: truthful faces and lying faces. Faces flashed for 100ms and then the participants rated them. However, the limitations of this study would be that it only had 15 participants and the mean age was 24.
John Kircher, Doug Hacker, Anne Cook, Dan Woltz and David Raskin have developed eye-tracking technology at the University of Utah that they consider a polygraph alternative. This is not an emotional reaction like the polygraph and other methods but rather a cognitive reaction. This technology measures pupil dilation, response time, reading and rereading time, and errors. Data is recorded while subjects answer true false questions on a computer.
They have found that more effort is required by lying than giving the truth and thus their aim to find indications of hard work. Individuals not telling the truth might, for instance, have dilated pupils while also taking longer to answer the question.
Eye-tracking offers several benefits over the polygraph: lower cost, 1/5th of the time to conduct, subjects do not need to be “hooked up” to anything, and it does not require qualified polygraph examiners to give the test.
Voice stress analysis
Voice stress analysis (also called voice risk analysis) uses computers to compare pitch, frequency, intensity and micro tremors. In this way voice analysis “detect[s] minute variations in the voice thought to signal lying.” It can even be used covertly over the phone, and has been used by banking and insurance companies as well as the government of the United Kingdom. Customers are assessed for truth in certain situations by banks and insurance companies where computers are used to record responses. Software then compares control questions to relevant questions assessed for deception. However, its reliability has been debated by peer-reviewed journals. “When a person lies, an involuntary interference of the nerves causes the vocal cords to produce a distorted sound wave, namely a frequency level which is different from the one produced by the same person when telling the truth.”
Several studies published in peer reviewed journals showed VSA to perform at chance level when it comes to detecting deception. Horvath, McCloughan, Weatherman, and Slowik, (2013), for example, tested VSA on the recordings of interrogation of 74 suspects. Eighteen of these suspects later confessed, making the deception the most likely ground truth. With 48% accurate classification, VSA performed at chance level. Several other studies showed similar results (Damphousse, 2008; Harnsberger, Hollien, Martin, & Hollien, 2009). In 2003, the National Research Council concluded “Overall, this research and the few controlled tests conducted over the past decade offer little or no scientific basis for the use of the computer voice stress analyser or similar voice measurement instruments”.
Functional magnetic resonance imaging is a technique used for multiple purposes which shows the uses of oxygen by the brain, allowing for the identification of which portions of the brain are using more oxygen, and thus being used during a specific task. This is called the Blood Oxygen Level Dependent or BOLD hemodynamic response. The first model of the magnetic resonance imaging (MRI) was built by Raymond Damadian and his colleagues in 1976. It revolutionized the field of anatomical study by providing images in real-time and 3-D models of human parts. The technique is also used in drug development, a wide-variety of research efforts, and diagnostically.
Studies using functional magnetic resonance imaging (fMRI) have shown that it has potential to be used as a method of lie detection. While a polygraph detects changes in activity in the peripheral nervous system, fMRI has the potential to catch the lie at the ‘source’. To use an MRI as a lie detector, an fMRI should be used by placing a magnetic band as a scanner on a subject’s head. However, the neurobiological systems that relate to lying are currently poorly understood. The current consensus is that faced with a forced choice paradigm, in which a subject has the choice of telling the truth or spontaneously generating a lie, lying can be distinguished due to increased prefrontal and parietal lobe activity. More specifically, the superior medial and inferolateral prefrontal cortices show net activation in the process of spontaneous lie generation (which involves suppression of the truthful response as well as generating a conceivable lie). There also is evidence of increased activation in the anterior cingulate cortex when lies are told. The fMRI shows the use of oxygen by the brain, allowing for the identification of which portions of the brain are using more oxygen during a specific task. By studying the brain images, researchers are able to map the systematic procedure the brain went through to produce the action or decision. Subjects are often offered monetary incentive if they can successfully deceive the process in hopes of generating a ‘real world’ scenario. Using this method, an initial 2005 study on individuals ( not group averages as previous studies) without pattern recognition and automation showed that lies can be distinguished 78% of the time. That statistic has risen, in one study, to 100% when predicting a lie in an individual when baseline lie/truth levels were closely studied with training from pattern recognition technology (machine learning). fMRI does rely upon the individual remaining still and safeguards in the analysis such that the questions can not be gamed by the participant (G. Ganis 2010). Studies have been done on Chinese individuals and their language and cultural differences did not change results. To show the robustness of this fMRI technology, a study (S. Spence 2011) was done that showed fMRI lie detection / truth verification technology worked even in a group of 52 schizophrenic patients, 27 of whom were experiencing delusions at the time of the study.
Attempts to introduce fMRI lie detection evidence in US Federal and State courts have been unsuccessful. In 2007 on episode 93 of the TV program MythBusters, the three members of the build team attempted to fool a non-automated fMRI test done by Cephos Corporation. Although two of the members were unsuccessful at fooling Cephos, the third member was able to successfully fool Steve Laken of Cephos, according to this member by keeping his mind in constant activity. The one out of three failure rate suggested that fMRI-based lie detection required further development. Also in 2007, the University of Pennsylvania group, used fMRI to test a Washington Post journalist Joel Garreau. The reporter was asked to pretend to apply for a job using a resume that included among other true items, three specific biographical items that were false. The test was able to detect two out of three items the reporter lied about. In 2016, the University of Pennsylvania group found that under similar experimental conditions, fMRI was 21% more likely to detect a lie than the polygraph.
Main article: Functional near-infrared spectroscopy
Functional near-infrared spectroscopy also detects oxygen and activity in the brain like the fMRI, but instead it looks at blood oxygen levels. It is advantageous to the fMRI because it is portable, however its image resolution is of less quality than the fMRI.
Electroencephalography is used to detect changes in brain waves.
Cognitive chronometry, or the measurement of the time taken to perform mental operations, can be used to distinguish lying from truth-telling. One recent instrument using cognitive chronometry for this purpose is the timed antagonistic response alethiometer, or TARA.
People often evaluate lies based on non-verbal behavior, but are quick to place too much merit in misleading indicators, such as: avoidance of eye contact, increased pauses between statements, and excessive movements originating from the hands or feet. Non-invasive lie detection using non-verbal behavior is performed by the Silent Talker Lie Detector. Silent Talker monitors large numbers of microexpressions over time slots and encodes them into large vectors which are classified as showing truthful or deceptive behavior by artificial intelligence or statistical classifiers. Silent Talker research has been peer-reviewed in the Journal of Applied Cognitive Psychology and in the Journal of Neural Computing and Applications. The architecture was invented between 2000 and 2002 by a team at Manchester Metropolitan University.
Another computer-based system for detecting spontaneous microexpression is being developed at Oxford and the University of Oulu by teams directed by Tomas Pfister. Traditionally, micro-expressions are very difficult to recognize through automated facial expression analysis because of their short duration and involuntariness. Their short duration means only a very limited number of frames are available for analysis using a standard 25fps camera and their involuntariness means eliciting a particular expression to add to a comprehensive training database requires considerable time and psychological insights. To counteract the short-video length, the team used temporal interpolation to achieve a higher number of frames and Multiple Kernel Learning to improve classification. The end result was the Spontaneous Micro-expression Corpus (SMIC), which consists of 77 micro-expressions taken from 6 subjects. This is the first system that is able to recognize spontaneous facial micro-expressions with reliable accuracy, approximately 70% compared to the 50% by trained human analysts. As such, it will be a valuable tool for future computer vision studies geared towards automating the process of lie detection.
Main article: Truth serum
Truth drugs such as sodium thiopental, ethanol, and cannabis (historically speaking) are used for the purposes of obtaining accurate information from an unwilling subject. Information obtained by publicly disclosed truth drugs has been shown to be highly unreliable, with subjects apparently freely mixing fact and fantasy. Much of the claimed effect relies on the belief of the subjects that they cannot tell a lie while under the influence of the drug.
- Daniel D. Langleben
- Ecological fallacy
- Forensics in antiquity
- Silent Talker Lie Detector
- Thought identification
- Voice risk analysis