Subliminal salience search illustrated: EEG identity and deception detection on the fringe of awareness

Abstract

We propose a novel deception detection system based on Rapid Serial Visual Presentation (RSVP). One motivation for the new method is to present stimuli on the fringe of awareness, such that it is more difficult for deceivers to confound the deception test using countermeasures. The proposed system is able to detect identity deception (by using the first names of participants) with a 100% hit rate (at an alpha level of 0.05). To achieve this, we extended the classic Event-Related Potential (ERP) techniques (such as peak-to-peak) by applying Randomisation, a form of Monte Carlo resampling, which we used to detect deception at an individual level. In order to make the deployment of the system simple and rapid, we utilised data from three electrodes only: Fz, Cz and Pz. We then combined data from the three electrodes using Fisher's method so that each participant was assigned a single p-value, which represents the combined probability that a specific participant was being deceptive. We also present subliminal salience search as a general method to determine what participants find salient by detecting breakthrough into conscious awareness using EEG.

Figure 1. Pz grand average.

Grand average at Pz, for all conditions. Positive is plotted down; y-axis is microvolts; x-axis is milliseconds. Vertical dashed lines mark the P3b bounding window. The amplitude of Probe and Fake is clearly larger than the amplitude of the two Irrelevant conditions. Moreover, the peak of Probe is earlier than the peak of Fake.

Figure 2. Fz and Cz grand averages.

P3a grand averages at Fz and Cz, positive down, vertical dashed lines mark the region in which we search for the P3a (i.e. the bounding window). There is evidence of a large P3a for the Probe condition. Probe is also earlier than Fake.

Figure 3. Example stimuli.

List of example names, formatted as stimuli. Note that name 3 would not be shown immediately after name 4 as they have 2 letters (‘A’ and ‘L’) in the same position.

Figure 4. Selection of P3b-Pz null hypothesis distributions.

Illustrative P3b randomised null hypothesis distributions for three participants (1, 2 and 8, whose ERPs are shown in Figure 10). The true observed value is marked by a vertical line, with area above that line, which gives p-value, marked. Data from Participant 1, whose P3b effect is weakest (as seen in their ERP) produces a large p-value. On the other hand, the true observed value for Subject 8, the participant whose effect is strongest, falls far outside of the randomised null hypothesis distribution, resulting in a p-value<0.001.

Figure 5. Early vs. Late grand averages for all channels.

Grand average ERPs for Early vs. Late. The ‘Early’ trace is the grand average for the first half (in chronological order) of Irrelevant2 trials that each participant was presented, while the ‘Late’ trace is the grand average for the second half of Irrelevant2 trials. There is no clear indication of a difference between the two conditions (especially, the amplitudes of the ‘Late’ traces are not greater than ‘Early’ ones). Vertical dashed lines demarcate the relevant P3 bounding window.

Figure 6. Latency difference null hypothesis distributions.

Randomisation inferred null hypothesis distributions aggregated across all participants for latency differences between Fake and Probe. Black vertical lines mark true observed value and p-value region. P3a calculated on Fz electrode and P3b at Pz. These show that the difference in latency between Fake and Probe is significant.

Figure 7. Distribution of p-values obtained from the intrinsic validity test.

Distribution of p-values combined under Fisher's method obtained from the intrinsic validity false alarm testing procedure. The 15 different bars in each bin represent the data obtained from the 15 participants. As expected, the distribution of p-values is uniform.

Figure 8. Fz ERPs for all participants.

Fz ERPs for all participants; positive down. Dashed vertical lines represent the P3a bounding window. The bold line is the ERP for the Probe condition, while the thinner line is the ERP for the Irrelevant1 condition. The number on the top left of each plot indicates the participant. The P3a effect (Probe more positive early and/or more negative late than Irrelevant1) is identifiable for each participant.

Figure 9. Selection of P3a-Fz and P3a-Cz null hypothesis distributions.

Fz and Cz P3a (peak-to-peak) null hypothesis distributions for three representative participants (participants 3, 11 and 14). True observed value and Type I error region are marked in black.

Figure 10. Selection of Pz ERPs.

Pz ERPs for all conditions for the participants considered in Figure 4 (1, 2 and 8). Positive is plotted down. Vertical dashed lines mark the P3b bounding window.

Figure 11. Selection of Fisher score distributions.

Fisher (P3a-Fz, P3a-Cz, P3b-Pz combined) null hypothesis distributions for three participants (subjects 3, 11 and 14; see also Figure 9) using Fisher's method. Black vertical lines mark true observed Fisher scores and p-value regions.