Unveiling Deceptive Intentions: Insights from Eye Movements and Pupil Size

Recognition of deceptive intentions from the eyes has been of appealing interest in the last decades but is still unresolved. Here, we report the development of a paradigm based on the Concealed Information Test enabling the study of various kinds of deception, that is, faking and concealing. Based on a card game, we compared fixation as well as pupil behaviour while participants were instructed to fake, conceal, or tell the truth. We realized two different layouts of stimulus presentation. Fixations differed between concealing versus faking and telling the truth; pupil size additionally unveiled the object of deception. We infer that various kinds of deceptive behaviour must be carefully distinguished, and contribute to how to use gaze measures as indicators of deceptive intentions.


INTRODUCTION
For generations, the enduring story of Pinocchio has become a symbol of the subtle, yet often revealing, manifestations of deceit as his nose grew longer with each untruth.In the absence of miraculous body transformations, eyes have long been considered to reflect emotions, thoughts, and even hidden truths.Thus, evolving from the first lie detectors based on physiological reactions as cardiovascular, respiration, and electrodermal activity [19], an important field of research has looked into the detection of deceptive behaviours using the eyes which benefit from a remove tracking of physiology.So far, respective research assessed fixations, their frequency, location, and duration, as well as the motion of the pupil.However, the impact and even the direction of respective effects is still unclear.We developed a modified Concealed Information Test enabling the study of various kinds of deception, i.e. faking and concealing the truth, and compared eye movements in these two tasks to a task of telling the truth.Both, fixations and pupil motion, were informative about deceptive intentions but also differed between faking and hiding the truth.Two different stimuli layouts were shown to severely affect fixations behaviours.We infer that at least two kinds of deceptive intentions must be carefully distinguished, constituting a base to a novel understanding of deceptive behaviours.

Fixations as indicator of deceptive behavior
Studies investigating eyes as indicators of deceptive intentions often report fixations frequency and duration as indicating changes in cognitive load associated with dishonest behaviours [5].Some studies used eye tracking to ensure the veracity of people's responses to personality tests -still widely used for personnel selection or detection of deviant behaviour -since being able to fake these tests questions their legitimacy and usability.Deceptive intention was associated with a lower number of fixations and a lower dwell time on items, but a higher number of fixations located toward extreme or congruent responses [21,25].Recognizing deceptive intentions is also of interest for assessing the veracity of witnesses and suspects' knowledge during a crime investigation.Asking a guilty and an innocent group to answer a questionnaire about a mock crime, [4] also found a lower fixation time on statements when they were answered deceptively.Fewer eye movements in the lie tellers' group during an interview were also found when presenting participants with real crime videos and asking them to be honest or to fake their answers [28].Other studies also used eye tracking to detect deceptive intentions when trying to hide the recognition of a known face and associated this behaviour with fewer fixations on the known face [14].However, even though deceptive behaviours are traditionally believed to go along with higher cognitive load [5,20], the lower number of fixations and lower dwell time can also support that being deceptive in lab conditions was less cognitively demanding than responding honestly [9,25].Moreover, there are studies reporting contradictory findings: [6] as well as [22] observed an increase in fixation counts on items with deceptive behaviours in the context of personality tests.[17] reported longer fixations on crime details during mock crime investigation.Taken together, effects of deceptive behaviour on the frequency and duration of fixations are still unclear.

Pupil motion as indicator of deceptive behavior
Pupil size has become of very interest in detecting deceptive behaviour since it is believed to reflect changes in emotional load associated with dishonest behaviours [2,3] as well as cognitive load [5].Larger pupil dilation in deceptive groups compared to honest ones was observed in the presentation of crime evidence related to a mock crime [12], answering statements during crime investigation [4], or filling rating scales of questionnaires [6].Pupil size has also been investigated in a revisit of the well-known rock-paper-scissors game [15].In this game, each of the signs (rock-paper-scissors) beats one sign and is beaten by another.Thus, players do not want their opponent to know which sign they are going to choose so they try to hide any cues that would reveal it.When displaying the three signs one after another, a higher pupil dilation was observed when the chosen sign was being displayed.This change in pupil size was then used by other participants to recognize the chosen sign.However, as for fixations, also for mean pupil size there are inconsistent findings: curiously, lower pupil dilation in the lie tellers' group than in the truth tellers' group was observed when responding to open questions in an interview [28].The authors suggested that this result may come from a baseline definition problem since they used real crime videos as stimuli, potentially inducing a ceiling effect that might have blurred effects of deceptive intention.

The present study
It appears that deceptive behaviours are supported to induce either fewer or higher fixations, as well as larger or smaller pupil size.At this stage, it could be advisable to point out that while there is still no strong consensus on the effect direction, there seems also to be no consensus in the literature on the denomination to use to talk about these deceptive behaviours.Most studies appear to arbitrarily use the terms "faking", "lying", "deceit", "misleading" or any other name without clearly stating the choice of one term compared to another [8].This lack of clear denomination could be a source of a misunderstanding as to which kind of deceptive behaviour we actually refer to.Thus, in this study, we refer to the denomination "deceptive intention" as a generic term for intentional response distortion.In this generic class of deceptive intentions, there are at least two that can be distinguished; concealing information and faking information.
Thus, the goal of this present study is to develop a controlled experimental design to investigate different kinds of deceptive intentions using eye tracking.By doing so, we expect to get insight into what deception is by distinguishing at least two kinds of deception and learning how they could impact eye behaviours.Inspired by previous studies that used modified Concealed Information Test ( [26] for a comprehensive review) to detect deceptive intentions from the eyes [12,14], we decided to create a similar task using cards with numbers because of their simple layout that allows for easy target/distractor discrimination [24].In this study, participants were asked to choose a number between 1 to 6 on a card and were then presented with all the card possibilities on a screen.They were told that an algorithm would read their mind by looking at their eyes.Their instruction was either to tell the truth about the value of the card, to conceal it, or to fake it by trying to make the algorithm believe that they chose another card.By doing so, we expect to differentiate two kinds of deceptive intentions and to compare them with an honest condition as control.We tested two different designs of stimuli presentation.In our "6-cards design", all the six cards were displayed on the screen at the same time after the choice of the participant, whereas in our "1-card design", the six cards were displayed one after another on the screen.Except for this difference, the task was identical for the two designs.We analysed both fixations as well as pupil motion in an eyes-only paradigm.
The first goal of this study was to detect differences in fixations and pupil size when participants are concealing or faking information compared to telling the truth about this information, and to quantify the nature, direction, and effect size of these changes.The second goal of this study was to examine fixations and pupil size when participants were presented with the card they chose compared to the card they wanted to fake or another card.We refer to both goals with the following questions: Q1. How do fixations and pupil size differ when someone is faking, concealing, or telling the truth?Q2.For each instruction of faking, concealing, telling the truth, how do fixations and pupil size differ when someone is presented with the card they chose, the card they wanted to fake, or another card?

Design
Participants were instructed to imagine playing a homemade card game with a partner, an opponent, and a referee, which were respectively represented on a computer by a partner smiley, an opponent smiley, and a referee smiley.In this game, participants faced six UNO cards with numbers from 1 to 6 in front of them on a table and had to choose a number on one of them.After the choice of the number -called target -, participants were instructed about the interlocutor they were going to interact with, and then copies of the six UNO cards were displayed on a screen for 30s.The goal for the participants was to interact with the smileys about the target just by using their eyes.These smileys were told to be run by an algorithm that will try to read participant's eyes.There were three kinds of instructions: (a) In the 6-cards design, all the cards are presented at the same time on the screen for 30s.In the 1-card design, each card is presented one after another for 5s.For both designs, the appearance of the number cards is preceded by the choice of the target, the interaction instruction with the associated smiley, and the appearance of block cards at the same locations as the subsequent numbers.
(b) From left to right, the smileys presented during the interaction instruction represent the partner, the referee, and the opponent.-"Telling the truth' is the interaction with the partner smiley: they have to mentally inform it about the target.For instance, if they chose the number '2', they want the friendly smiley to recognize that they chose this number.
-'Faking the truth' is the interaction with the opponent smiley: they have to make it guess another card than the target.For instance, if they chose the number '2', they want the opponent smiley to recognize that they chose another number, like the number '5'.The choice of this faked number was up to the participant.
-'Concealing the truth' is the interaction with the referee smiley: they have to prevent it from guessing the target.For instance, if they chose the number '2', they do not want the neutral smiley to recognize that they chose this number.
The experiment was separated into two parts in order to test deceptive intention recognition in different designs.We refer to these parts as "6-cards design" and "1-card design".In the 6-cards design, the six cards appeared at the same time on the screen at six predefined positions for 30s (top-left, top center, top right, bottom left, bottom center, bottom right).The attribution of the cards to the screen position was randomized for each trial.In the 1-card design, the six cards were presented one after another in the middle of the screen for 5s each.The order of appearance of the cards was randomized for each trial.The card sequence for the two parts is illustrated in Fig. 1.a.The study followed a within-subject design: all the participants participated in both parts, but the order of presentation was reversed for half of them.

Apparatus and Stimuli
Eye movements were measured using the eye tracker EyeLink 1000 from SR Research Ltd.The setup consisted of one table supporting a screen and the eye tracker.Participants were sat at a distance of 57cm from a screen in a room with a low ambient light.Since we wanted to monitor for small variations of the pupil that could testimony for faking intentions, they had to put their head on a chin rest to prevent noise movements that could perturb video analysis.Instructions were white-coloured and presented against a grey background on a 54" display monitor with a resolution of 1920 x 1080 pixels and a 100Hz refresh rate.Stimuli presentation and response collection were performed using a custom program written on Psychopy 2022.2.5 [16], running on Windows 10, 64-bits.
Real blue UNO cards were used when participants had to choose the cards at the beginning of a trial.We chose to use these cards because of their simple and clear design that allows direct identification of the number.Minimizing the differences in brightness distribution in the stimuli, arranged numeric UNO cards were used for displaying the cards on the screen.Stimuli were created by numeric replicas transposed in grey and white colour with simple pixel discrimination.Stimuli's brightness was controlled among all the cards in order to present the same brightness.The numeric cards had a size of 5.9 x 4.1 degrees of visual angle.The area of interest associated with a card was defined with an additional degree of visual angle around the borders of the card in order to safely consider the Eyelink 1000 spatial accuracy, defining a rectangle of 7.9 x 6.1 degrees of visual angle.Smileys were used to represent the interlocutors when instructing participants about their interaction condition and were never part of eye-tracking recordings.A green friendly smiley represented the partner, an orange angry smiley represented the opponent, and a yellow neutral smiley represented the referee, all with a circular diameter of 5.0 degrees of visual angle and illustrated in Fig. 1.b.

Procedure
The experiment was approved by the Ethics Committee of the university.All participants were informed about the possibility of aborting the experiment at any moment and gave their written consent to participate in the experiment.After the presentation of the eye tracking system, participants were instructed about the task of the study as stated in the Design section.They were also instructed about the design of the physical cards and the numeric replica, as well as the colour and shape of the three smileys.The experiment was separated into two parts -6-cards design and 1-card design -that would last for approximately 40 minutes each.The order of presentation was reversed for half of the participants, but they all completed both parts.Each part was separated into three identical series to allow participants to have breaks to rest their eyes.A longer break was announced between the two parts.The series started with a calibration to allow the eye tracking system to associate the eye movements with the location of what the participant is looking at.During this calibration, participants had to follow with their eyes a dot moving randomly from the center to six different positions on the borders of the screen and then coming back to the center of the screen.The calibration phase lasted for approximately one minute and trials started once the calibration was done successfully.A series was composed of 12 trials, where each interaction instruction appeared 4 times in a random order.The six physical number cards were disposed next to each other on the table in front of the participant during the whole experiment to provide a visual context.The order of the cards was randomized and shuffled for every participant and trial series.Each trial began with the choice of a new target.Using the keyboard, participants were also asked to write down the value of the target before and after the display of the cards on the screen to make sure they remembered it.In the "Faking the truth" condition, participants were also asked about the faked card they wanted the opponent smiley to guess.Then, UNO block cards with neutral symbols were displayed on the screen for 5s before the display of the number cards in order to not make the appearance of the numbers induce a brightness reaction.Those block cards were located at the same place as the subsequent number cards: six block cards at the top-left, top center, top right, bottom left, bottom center, and bottom right positions for the 6-cards design, and one block card at the center of the screen for the 1-card design.Participants were told to communicate only using their eyes to reduce the amount of body movements in the experiment.Even if participants were instructed on "what" to do, they were not instructed on "how" to do it, which means that they had to develop their strategies in order to convey the proper information to their interlocutor.The only requirements were to look at the screen and not to move the body.

Participants
The study was advertised using the local university study management system (SONA Systems) and participants received hours credits as compensation for their participation.The requirements for participation was normal or corrected-to-normal vision and the absence of epileptic seizures.A total of 22 participants, including 13 females and 9 males, were recruited in the study (M=22.05,sd=3.05).

RESULTS
An error rate was defined by the proportion of trials where there was a mismatch between the first call of the target before the display of the cards on the screen and the second call after the presentation of the cards.The mean error rate across participants was 5% with a maximum of 12.5%.No participant was excluded based on this error rate.Trials that contained a mismatch were however excluded.Gaze data were exported using the EyeLink Data Viewer software package [23].The software extracts the horizontal and vertical gaze coordinates from the tracker to classify fixations.The raw data to analyse the pupil was extracted from the EyeLink software sample reports.The analysis was performed on R [18].Linear mixed models were used for the analysis of fixations frequency and duration and pupil size, using the LME4 R package [1].F-values and p-values were then estimated using Satterthwaite's method.We used Tukey's difference test for post hoc analyses to further explore the interactions between factors.A post hoc power analysis was performed based on [10] methodology to investigate the reliability of our results, showing sufficient reliability of at least 0.76 for each variable.The associated chart can be accessed in the appendixes section.

Fixations
Fixations shorter than 60ms were removed from the datasets.Fixations outside the screen boundaries were also removed due to their lack of trustworthiness, as the eye tracker was calibrated on the screen.To investigate the effect of the instruction condition (Q1), the total number of fixations and the mean duration of fixations on the screen were entered in linear mixed models performed respectively on the frequency and duration of fixations.Deceptive instruction was entered as a fixed effect and participants' ID as a random effect.To investigate the effect of the type of stimulus for each one of the instruction conditions (Q2), the total number of fixations and the mean duration of fixations on the card's areas of interest were entered in linear mixed models performed respectively on frequency and duration of fixations.Deceptive instruction and the stimulus type were both entered as fixed effects and participants' ID as a random effect.No participants were excluded for fixations analysis.
Q1. How do fixations differ when someone is faking, concealing, or telling the truth?Q2.For each instruction of faking, concealing, telling the truth, how do fixations differ when someone is presented with the card they chose, the card they wanted to fake, or another card?Number of fixations.The number of fixations on the stimuli considered for this question was divided by the number of stimulus types on the screen: 5 distractors were on the screen for the Telling and the Concealing instructions, and 4 distractors were on the screen for the Faking condition due to the presence of the faked target.The interaction effect of the instruction and the stimulus type on the number of fixations was significant for the 6-cards design (F=21.3,p<2e-8) (see Fig. target (M=99, diff=63.0,p<0.0001) for the Faking instruction.No significant differences were found between target and distractors for the Faking instruction.The interaction effect of the instruction and the stimulus type on the number of fixations was also significant for the 1-card design (F=4.9, p<0.01) (see Fig. 3.b.left).A higher number of fixations were located on the target (M=98.9)compared to distractors (M=71.1,diff=27.8,p<0.02) for the Telling instruction.No significant differences were found between target and distractors for the Concealing instruction.A higher number of fixations were located on the faked target (M=93.9)compared to distractors (M=68.8,diff=25.1,p<0.03) and target (M=65.3,diff=28.6,p<0.008) for the Faking instruction.No significant differences were found between target and distractors for the Faking instruction.
Duration of fixations.The interaction effect of the instruction and the stimulus type on the fixations durations was significant for the 6-cards design (F=27.5, p<2e-10) (see Fig. 3.a.right).Lower fixation durations were spent on the target (M=224.2ms)compared to distractors (M=616.2ms,diff=392.0ms,p<0.0001) for the Telling instruction.No significant differences were found between target and distractors for the Concealing instruction.Lower fixation durations were spent on the faked target (M=314.1ms)compared to distractors (M=545.4ms,diff=231.3ms,p<0.0001) and target (M=529.4ms,diff=215.3ms,p<0.0001) for the Faking instruction.No significant differences were found between target and distractors for the Faking instruction.For the 1-card design, the interaction effect of the instruction and the stimulus type on the fixation's durations was not significant (F=1.2,p>0.31) (see Fig. 3.b.right).

Pupil motion
The 1-card design followed [13] recommendations for experimental design involving pupillometry.Stimuli were presented one at a time on the same position at the center of the screen to induce a constant or close to constant eye position.The presentation of the successive stimuli was slowpaced with a 5s delay to make sure the stimuli's reactions were not overlapping.A card was always presented on screen in order to maintain a similar visual environment during the trial.Stimuli's brightness was controlled to have a global brightness under 1% of difference, and a similar distribution of grey and white pixels.Even though the 6-cards design was less suited for analyzing pupil motion for each stimulus due to the presentation of multiple stimuli at the same time, we were still able to investigate mean differences in pupil size between the deceptive instructions (Q1), but not regarding the stimulus type (Q2).
Pupil data was extracted from EyeLink software sample reports for each participant.Preprocessing and cleaning of pupil data were performed using the PupillometryR package [7].Trials with more than 20% of missing data were removed for analysis.Blinks were defined by missing data and were removed for the analysis, as well as the 100ms before and after the blinks.Gazes outside of the screen were also removed.After the definition of the baselines, data was down-sampled from 1000Hz to 100Hz by computing the mean pupil size for 10ms bins to save computing power and grasp significant size changes.After cleaning, two participants still had more than 50% of their gaze outside the area of interest dedicated to the card, while the gaze data on cards didn't allow computation of mean pupil size for more than 15% of the time.These two participants were removed from the dataset due to the lack of trustworthiness of their pupil measures, leading to a final cohort of 20 participants for pupil analysis.
Q1. How does pupil size differ when someone is faking, concealing, or telling the truth?For the 1-card design, baselines were defined by the 50ms before a number card appeared on the screen.Thus, a new baseline was defined every 5s to consider the evolution of the pupil size in regard to each stimulus.Six baselines were defined for the presentation of the six cards of a trial.However, correcting the pupil size for the appearance of each card did not allow the comparison of mean pupil size in the different deceptive instructions.Thus, we also computed the not-corrected mean pupil size for each baseline group and each participant and performed a linear mixed model with deceptive instructions and baseline groups as fixed effects and participants' ID as a random effect (see Fig. 4.a).The units are displayed in arbitrary units.The effect of the deceptive instruction on mean pupil size was significant (F=3.7,p<0.03).We were interested to know in which baseline group this difference was coming from and performed linear mixed models for each baseline group.The effect of the deceptive instruction on mean pupil size was significant for the first baseline group (F=5.0,p<0.02).For this group, the mean pupil size was significantly higher for the Faking instruction than for the Telling instruction (diff=78.9A.U, p<0.009).No significant differences were found between the Faking and the Concealing instructions or between the Concealing and the Telling instructions for the first baseline group.No significant differences were found for the other baseline groups.To help the understanding of the evolution of pupil size with time in the different conditions, we did simple linear regressions of the mean pupil size across baseline groups for each instruction condition.These regressions revealed stronger decreasing slopes for the Faking (y a*x+b, a=1919, b= -46.4,adj.R2 = 0.72) and the Concealing instructions (y a*x+b, a=1881, b= -45.2, adj.R2 = 0.73) compared to the Telling instruction (y a*x+b, a=1841, b= -30.0, adj.R2 = 0.68), suggesting that the pupil size decreased faster for the two deceptive instructions.
To have the same analysis methodology between the two designs, size groups were defined in the same way in the 6-cards design, i.e. by computing 50ms mean pupil size every 5s, and starting 50ms before the display of the six cards on the screen.The effect of the deceptive instruction on mean pupil size was not significant.The effect of the deceptive instruction on mean pupil size was not significant for the baseline groups (see Fig. 4.b).However, we can still pinpoint a marginal significance for the first baseline group (F=2.9, p<0.07).For this group, the mean pupil size was significantly higher for the Concealing instruction than for the Telling instruction (diff=78.9A.U, p<0.009).No significant differences were found between the Faking and the Concealing instructions or between the Concealing and the Telling instructions for the first baseline group.Linear regressions of the mean pupil size across baseline groups for each instruction condition revealed stronger decreasing slopes for the Faking (y a*x+b, a=1807, b= -53.6, adj.R2 = 0.89) and the Concealing instructions (y a*x+b, a=1791, b= -47.3, adj.R2 = 0.66) compared to the Telling instruction (y a*x+b, a=1749, b= -37.8, adj.R2 = 0.67), suggesting that the pupil size decreased faster for the two deceptive instructions.However, a post hoc power analysis revealed that the results concerning the effect of instructions on the mean pupil size may be underpowered (0.76, see Appendix) and should therefore be considered with caution.
Q2.For each instruction of faking, concealing, telling the truth, how does pupil size differ when someone is presented with the card they chose, the card they wanted to fake, or another card?The 6-cards design was less suited for analysing pupil motion specifically for each stimulus type due to the presentation of multiple stimuli at the same time on the screen.Thus, only the gazes on the card in the 1-card design were considered to investigate the evolution of pupil size in regard to the stimulus type for each instruction condition.As described in the previous part, pupil size was corrected by a baseline defined by 50ms before the appearance of a number card on the screen.For each deceptive instruction, linear mixed models were performed on the baseline-corrected pupil size with stimulus type as a fixed effect and participants' ID as a random effect.
For the Telling instruction, the mean pupil size was significantly higher when presenting the target (M=1.005)compared to a distractor (M=0.991)(F=302.7,diff=1.4%,p<0.001).When looking at the time evolution of the pupil size, we can see that for both stimulus types, the presentation of a card was associated with a positive peak in pupil size around 750ms after stimulus onset.This peak was higher for the presentation of the target with a dilation of about 2.9%, compared to 0.8% for the presentation of a distractor.A second peak occurred only for the target around 3200ms after stimulus onset with a dilation of about 1%.The results are presented in Fig. 5.a.
For the Concealing instruction, the mean pupil size was significantly higher when presenting the target (M=1.001)compared to a distractor (M=0.987)(F=483.0,diff=1.4%,p<0.001).When looking at the time evolution of the pupil size, we can see that the presentation of the target was associated with a positive peak in pupil size around 1500ms after stimulus onset, while no peak occurred for the presentation of a distractor.The dilation in pupil size for the target was about 2.3%.A second peak occurred only for the target around 3000ms after stimulus onset with a dilation of about 0.7%.The results are presented in Fig. 5.b.
For the Faking instruction, the effect of the stimulus type on mean pupil size was also significant (F=637.0).The mean pupil size was significantly higher when presenting the target (M=1.028)compared to a distractor (M=0.995,diff=3.3%,p<0.001) or the faked target (M=0.987,diff=4.1%,p<0.001).The mean pupil size was significantly higher when presenting a distractor compared to the faked target (diff=0.8%,p<0.001).When looking at the time evolution of the pupil size, we can see that for the three stimulus types, the presentation of a card was associated with a positive peak in pupil size around 700ms after stimulus onset.This peak was higher for the presentation of the target with a dilation of about 3.2%, compared to 1.9% for the presentation of the faked target and 0.6% for the presentation of a distractor.A second peak occurred only for the target around 2000ms after stimulus onset with a dilation of about 5%.The results are presented in Fig. 5.c.

DISCUSSION
In the current study, we tested a new version of a Concealed Information Test enabling us to investigate different deceptive behaviours in one experimental context.The first goal of the study was to detect and quantify any differences in fixations and pupil size when participants were instructed to conceal or fake information compared to telling the truth about it.We found different fixations and pupil behaviours in Concealing versus Faking instructions, suggesting that both deceptive intentions induce different effects on the eyes.The second goal of this study was to detect and quantify any differences in fixations and pupil size when participants were presented with the card they chose compared to another card, or the card they wanted to fake.Fixations indicated target for Telling and faked target for Faking, which were the objects that participants wanted to communicate about.Pupil size additionally revealed the target for both Concealing and Faking instructions even though participants had to hide it.These two questions were investigated in two different ways of presenting the stimuli: a 6-cards design where all the cards were presented at the same time on the screen after the definition of the target, and a 1-card design where all the cards were presented one after another at the center of the screen.These two designs induced different fixation behaviors despite the same task.
4.1 Q1.How do fixations and pupil size differ when someone is faking, concealing, or telling the truth?Our findings demonstrated that concealing and faking lead to different fixation and pupil behaviours, suggesting that they do not rely on the same deceptive processes.We can recognize when someone is trying to conceal information compared to being honest or fake information thanks to the mean number and the mean duration of fixations for both designs of stimuli presentation.Concealing in the 6-cards design involved more fixations than Telling or Faking, and went along with smaller fixations duration.On the opposite, Concealing in the 1-card design involved fewer fixations than Telling or Faking, along with higher fixations duration.Since the target was always on screen in the 6-cards design, this suggests that participants wanted to blur the recognition of the target in the Concealing condition by making more eye movements on all the cards.In the 1-card design, the target was not always on the screen and participants did not know when it would appear.Reducing the number of eye movements would have been a way to conceal the target in order to decrease potential variations in eye reactions to the presentation of the target.For both designs, the Concealing instruction induced different eye movements compared to the other instructions, suggesting a strong effect of this specific deceptive behaviour on the eyes independently of the stimuli distribution.This may be due to the need to communicate about a card in Telling and Faking instructions but not in the Concealing one.We were not able to recognize when someone was trying to fake a card compared to being honest with the mean number or mean duration of fixations, meaning that the Faking and Telling instructions induced similar eye movements.Previous literature supported that faking information induces a higher cognitive load than being honest [27], inducing a different fixation behaviour [28].As faking or being honest about the value of a card did not influence the mean number of fixations in our study, it seems like no differences in cognitive load were observed in gaze behaviours between Faking and Telling instructions.We suggest that the necessity to hide the real target in the Faking condition was not inducing enough changes in cognitive load to observe differences in eye movements compared to an honest condition.
Pupil size may be indicative of different deceptive intentions.The mean pupil size in the first time group, i.e. 50ms before the presentation of the first stimulus, was larger for both deceptive instructions compared to the honest one in both designs.The mean pupil size was not significantly different for the other time groups.This would mean that Faking, Concealing, and Telling instructions induced different mean pupil sizes just before presenting the cards on the screen, but then this effect flattened with time.We suggest that this difference may come from a different emotional reaction to the instruction: participants were more likely to be aroused when cards were going to be presented and they were instructed to act deceptively compared to acting honestly, resulting in a higher pupil size.Then this emotional load was regulated as the cards were presented.Moreover, linear regressions suggested a faster decrease in mean pupil size with time for the Faking and Concealing instructions than for the Telling instruction, as well as for Faking compared to Concealing.We may intuitively think that behaving honestly would be less exciting than behaving dishonestly, leading to flattened pupil changes in the honest condition.Thus, the slow pupil constriction may reflect a decline in arousal which would be more pronounced in both deceptive instructions than the honest one, resulting in a faster decrease in mean pupil size.A post hoc power analysis revealed that our results concerning the effect of instructions on the mean pupil size may be underpowered and should be considered with caution until further studies try to replicate these observations.
The findings related to Q1 suggest that Concealing information and Faking information are two different cognitive processes that do not lead to the same eye behaviours and could be distinguished by the number and duration of fixations, and the pupil size.Fixations differed for concealing versus faking and telling the truth, while mean pupil size could differentiate both deceptive instructions from the honest one at least during the first 5s of stimuli presentation.Considering that no previous study to our knowledge talked about this differentiation of deceptive intentions, we carefully suggest that the lack of accordance on which kind of deceptive intention we refer to could be a first explanation for the gap in results directions we may sometimes observe.Indeed, it seems like studies on deceptive behaviours arbitrarily choose to use the terms "faking", "lying", "deceit", or "misleading" to indicate the behaviours they investigate.In this study we identified two different deceptive intentions from the eyes, Faking and Concealing.If such results are confirmed in subsequent studies, we would advise using the term "deceptive" to refer to general deceptive behaviours, and to be precise about what kind of deceptive intention is investigated by at least distinguishing "concealing" intentions from "faking" intentions.
4.2 Q2.For each instruction of faking, concealing, telling the truth, how do fixations and pupil size differ when someone is presented with the card they chose, the card they wanted to fake, or another card?Fixations indicated the object that participants wanted to be known -target for Telling, faked target for Faking -while pupil size also unveiled the target for both Concealing and Faking even though participants were trying to hide it.We were able to recognize the target for the Telling instruction thanks to fewer and shorter eye movements compared to a distractor in the 6-cards design.On the contrary, the target was associated with a higher number of eye movements of similar duration compared to a distractor in the 1-card design.The presentation of all the cards on the screen at the same time for the 6-cards design enhanced the possibilities of eye movements.Thus, decreasing the amount of eye movements on the target compared to distractors may have been the most natural way to make it stand out.Moreover, the lower fixation duration on the target may reveal that it is cognitively easier to look at the card we want to communicate about compared to distractors.In the 1-card design, fewer possibilities of eye movements were expected due to the presence of a single card on the screen at a single location, so increasing the number of eye movements on the target compared to distractors may have been the most natural way to make it stand out.We found the same effects of eye movements associated with the presentation of the faked target for the Faking instruction than for the target for the Telling instruction, while no differences in eye movements were found between the target and distractors for both the Faking and Concealing conditions.Thus, eye movements did not allow the recognition of the real target in the deceptive conditions but only the recognition of the card participants wanted to communicate about, and the communication strategies appeared to be similar for the Telling and the Faking instructions, without being impacted by the existence of a faked target.
Pupil motion was indicative of the stimulus type for each instruction condition.For Telling, Concealing and Faking, the presentation of the target was associated with a higher pupil dilation than the presentation of a distractor.In the Faking condition, the pupil dilation subsequent to the presentation of the faked target was smaller than for the target but larger than for a distractor.This would mean that the presentation of the target induced a higher physiological response than the presentation of a distractor, even under deceptive instructions.This phenomena may arise from the creation of an emotional bond with the target due to the voluntary choice of this specific card.For Faking instruction, the presentation of the card supposed to be faked also induced a physiological reaction higher than a distractor, but lower than the real chosen card.Moreover, this first pupil dilation occurred around 700ms after stimulus onset for the Telling and the Faking instructions, and about 1500ms after stimulus onset for the Concealing instruction.Participants were trying to control their reaction to all the cards in the Concealing instruction, while they were supposed to communicate about a card in the Telling and Faking conditions.We may assume that this "control" mental state led to a slower pupil reaction to the card on screen for the Concealing condition.After the first pupil dilation, a second pupil dilation occurred for the presentation of the target.This second dilation occurred around 2000ms for the Telling and the Faking instructions, and around 3000ms for the Concealing instruction.Interestingly, this second dilation was smaller than the first one for Telling and Concealing instructions, but higher for the Faking instruction.If we can safely attribute the first pupil dilation to a physiological reaction to the content of the card, the origin of the second dilation should be further discussed.We assume that it may arise from other cognitive processes as the processing of the card content.However, we leave it to subsequent studies' hands to replicate and highlight the presence of this second pupil dilation before concluding strongly about it.
The findings related to Q2 suggest that eye movements may reveal strategies and patterns of communication about a specific card, but they hardly inform about the real card chosen by participants under deceptive instructions.However, pupil motion was able to catch a higher emotional reaction to the presentation of the target from distractors for both Faking and Concealing conditions, potentially coming from the pupil size being more spontaneous and less sensitive to voluntary control than eye movements [11].Thus, in a Concealed Information Test design, eye movements may be used to expose behaviours that are linked with the concealing of information compared to being honest about it, while pupil motion can be used to recognize the object of the deceptive act, whether it is the concealing or the faking of information.

Further considerations
In this study, we used two different designs of stimuli presentations to investigate deceptive intentions using eye tracking.All the cards were presented at the same time in the 6-cards design, and all the cards were presented one after another in the 1-card design.That means that participants knew that the target would be on screen from the beginning of the trial for the 6-cards design, while they did not know when the target would appear for the 1-card design.Both designs induced changes in fixations number and duration for the Concealing instruction compared to the Telling and Faking instructions.However, despite the same instruction, differences in the presentation of the stimuli induced opposite results for the frequency and duration of fixations.This suggests that the way the stimuli are presented can strongly influence participants' eye movement strategies according to the deceptive intention.We know that cognitive and emotional load can be used as leverage to increase deception recognition [3,5].Here, we support that the way of presenting the stimuli can also strongly influence the subsequent expression of deceptive behaviours in the eyes.Thus, further studies that try to detect deceptive behaviours using eye parameters should also care about the way the stimuli are presented to predict eye movements.
Several limitations have to be considered about the results of this study.Participants were instructed to lie or to tell the truth, which could question the validity of our results if participants were free to deceive the algorithms or not.There were also no consequences for being deceptive or honest with no fear of being caught.This entertaining environment may have induced a lower involvement of the participants in the tasks compared to being deceptive in situations with real consequences.Moreover, even though the use of cards in a laboratory environment allowed improved control and investigation of new parameters, this study was performed using a desktop eye tracker which required a chin rest.The participants were also told to only use their eyes to do the task.These restrictions may have impacted how the deceptive acts were naturally performed.The results of this study would also benefit from replications in more applied scenarios to extend their reliability.Also, our sample of participants is from a similar university environment with most of our participants being students of similar age and educational levels.These results may not be the reflection of a global population if we suspect changes in deceptive abilities due to age or educational background.The use of coloured and shaped smileys as visual objects to help the differentiation of the conditions may have also influenced gaze behaviours.Evaluating the impact of the colour and shape of instruction objects on eye movements and deceptive behaviours was beyond the scope of the study, but would be an interesting issue to be addressed by further studies.Finally, prediction models based on the findings of this study would greatly support the use of fixations and pupil size as indicators of deceptive intentions.Fixations and pupil size were both indicators of concealing intentions, while faking intentions were more difficult to detect.Therefore, we suspect that -although having just group differences at issue here -developing a prediction algorithm to detect concealing intentions from eye movements might be realizable in the near future.For faking intentions, unfortunately, more research seems to be necessary.Such