7.3 Results & discussion
8.4.1 Target difficulty affects the P3 for targets outside but not inside the
As shown in Section 8.3.1, there is a significant difference in accuracy scores between easy and hard targets both outside and inside the AB. Hence, in terms of behavioural accuracy, the identity of a target letter (i.e. whether it belongs to the easy or hard category of letters) has an influence on target detection both if the target is presented individually (i.e. outside the AB) and also if it is presented during the AB.
In Chapter 6, we showed that target difficulty affects the size of the P3 component for individually presented targets (i.e. targets presented outside the AB) that are correctly reported. We proposed that an easy target letter has more bottom-up strength than a hard letter and this increases the size of the P3 component evoked by the easy target. The results presented in Section 8.3.2 re-emphasise how, for targets outside the AB, ‘easy-hardness’ of targets affects the size of the P3 component.
In Section 8.3.2, we perform the same analysis for targets presented during the AB. From the behavioural analysis (Section 8.3.1), we know that intrinsic stimulus characteristics
(i.e. whether the target is easy or hard) affect target perception if the target is presented during the AB. Interestingly, however, ‘easy-hardness’ of targets does not influence the P3 component for targets during the AB. The bar chart in Figure 47B illustrates that there is no statistically significant difference in P3 size for easy and hard targets, both if the target is correctly and incorrectly reported. It seems that the P3 component is influenced by different factors depending on whether a target is presented outside or inside the AB, which is intriguing.
8.4.2 Virtual ERPs from the ST2 model
In the previous chapters of this thesis, we have shown how we can visualise the theory underlying the ST2 model by plotting virtual ERPs for the conditions of interest. Accord- ingly, we express the ST2 model’s theoretical standing on the effect of target strength on
target perception outside and inside the AB by generating the corresponding virtual ERPs. In the human data, we have no direct measure of target strength and, hence, use target difficulty (i.e. whether a target is easy or hard) as an indirect measure of target strength. In the model, however, target strength is precisely defined by each target’s strength value. Consequently, we generate virtual ERPimages (see Section 4.3.5 for a description of vir- tual ERPimage methodology) that are sorted by the target strength value and simulated accuracy on each trial.
Targets presented outside the AB
If a target is presented to the ST2 model individually (i.e. outside the AB), the blaster is available. Consequently, the blaster enhances the item’s type representation as soon as this item has been identified as a target by the task filter. Whether a target is successfully tokenised depends on its bottom-up strength. After receiving an enhancement from the blaster, targets outside the AB will normally have enough bottom-up strength for their type representation to bind to a token and thus they will be encoded into working memory. Very weak targets, however, fail to gather sufficient strength and cannot initiate tokenisation. Such items fail to bind to a working memory token and are not ‘seen’ by the ST2 model.
Figure 48 shows the activity underlying the virtual P3 for a single target in RSVP. The colour strips on the left indicate the target’s accuracy and strength, respectively, for the
Time (ms) −2001 0 200 400 600 800 2 3 4 5 6 7 8 9 10 11 12 ST2 Single target
Sorted by increasing target strength: low (blue) -> high (red) Target accuracy: red - incorrect; green - correct
Target vP3
Figure 48 Virtual P3 from the ST2 model for targets outside the AB (a single target in RSVP).
Colour strips to the left of the plots indicate the accuracy and target strength for that particular trial of the simulation.
corresponding trial of the simulation. The virtual ERPimage is sorted by target strength, from the lowest strength value at the bottom to the highest at the top of the plot. The virtual ERPimage shows how in the ST2model, targets have to overcome a critical strength value before they are able to initiate tokenisation. Whereas targets with strength values below the threshold are ‘missed’, targets above the threshold are ‘seen’ by the ST2 model.
Targets presented inside the AB
As discussed in Chapter 7, the duration of T1’s tokenisation is determined by T1’s strength. The blaster is suppressed while T1 is tokenised, hence, the availability of attention for T2 during the AB depends on how long it takes to tokenise T1. For targets presented inside the AB, successful tokenisation thus depends not only on T2’s strength, but also on the strength of the preceding T1 and consequently the availability of attention.
Figure 49 illustrates this issue in the virtual P3 ERPimage for a T2 presented at lag 3 following a correctly identified T1, i.e. a target inside the AB. Again, this plot displays T2 accuracy and strength to the left of the figure and trials are sorted by T2 target strength.
Time (ms) −200 0 200 400 600 800 20 40 60 80 100 120 140 ST2
Sorted by increasing target strength: low (blue) -> high (red) T2 accuracy: red - incorrect; green - correct
T2
T2 vP3
T1 vP3
Figure 49 Virtual P3 from the ST2 model for targets inside the AB, i.e. a T2 presented at lag 3
following a correctly reported T1. Colour strips to the left of the plots indicate the accuracy and target strength for that particular trial.
Furthermore, within each T2 strength value, the trials are sorted in ascending order by T1’s target strength values. In contrast to targets outside the AB, the blaster is suppressed by T1’s tokenisation process during the AB. As target representations decay over time, many targets with lower strength values (trials 1-80 in Figure 49) fail to ‘outlive’ the unavailability of attention and cannot be tokenised. Even targets with medium strength values (trials 60- 100 in Figure 49) are mostly ‘missed’ during the AB. These trials show some marginal virtual P3 activity, which is due to activity in higher layers of stage one contributing towards the virtual P3. Only a few medium strength targets, where T1 is tokenised particularly quickly and the blaster becomes available earlier, are able to initiate the tokenisation process and are ‘seen’ by the ST2 model. Targets with high strength values (displayed at the top end of the ERPimage in Figure 49) have sufficient bottom-up strength to be tokenised despite being presented during the AB.
The ST2 model fails to explain the results for targets during the AB
For targets outside the AB, the higher the target strength value, the greater the likelihood that the target will be correctly reported by the ST2 model. Hence, target strength directly affects simulated behavioural accuracy. Furthermore, virtual P3 size increases with larger target strength values. A target strength value is the model’s equivalent of a target’s intrinsic stimulus characteristics in the human data (i.e. whether it is easy or hard; see also Chapter 6). Consequently, in the ST2 model, whether a target is easy or hard influences both simulated accuracy and the virtual P3. For targets outside the AB, the model is thus in line with the human results.
In most cases, target detection during the AB depends on the availability of attention, which, in turn, is determined by the amount of time it takes to process the preceding target (i.e. the T1). Consequently, there are some trials where a stronger T2 is ‘missed’ because T1 processing takes too long and, vice versa, other trials where weaker T2s are ‘seen’ because T1 is tokenised quickly. Nevertheless, target detection and especially the virtual P3 for targets during the AB (see Figure 49) are also strongly influenced by target strength, i.e. targets with higher strength values generate larger virtual P3 components. This is in contrast with the human data from Section 8.3.2, where, during the AB, the P3 component seems unaffected by target strength.