EEG frequency analysis

Methods

As in Experiments 1 and 2, we applied the three frequency-based measures evoked power (EPow), whole power (WPow) and phase locking index (PLI) for the EEG analysis (cf.

Experiment 1). All measures were determined by Gabor wavelet analyses in frequency bins of 0.5 Hz (time window –334 to 1000 ms plus 50% tapering window). As for Experiment 2, analyses were confined to lower frequency bands (< 6 Hz) for the midline electrodes FZ, CZ, PZ and the parietal electrodes P3 and P4.²⁰ The statistical analysis was carried out as in Experiment 2.

Results

Figure 4.5 shows that, for both category violation conditions, there was a pronounced decrease of delta band activity (1-3 Hz) in comparison to the antonym condition for all three applied measures (EPow, WPow and PLI). This decrease was confined approximately to the time range of the N400 effect of the corresponding ERP analysis.

The statistical analyses confirmed these observations. For EPow, there was a significant main effect of TYPE for the electrodes PZ (F (2,32) = 20.10, p < .01) and P4 (F (2,32) = 22.18, p <

.01) between 100 and 500 ms (averaged frequency bins: 1.0-3.0 Hz). Single comparisons for each category violation condition in comparison to the antonym condition revealed significant differences for non-related category violations (PZ: F (1,16) = 24.29, p < .01; P4: F (1,16) = 23.28, p < .01) as well as for related category violations (PZ: F (1,16) = 21.92, p < .01; P4: F (1,16) = 25.77, p < .01), but no significant difference between them (PZ: F (1,16) = 1.46, p = .244; PZ: F < 1). This pattern emerged at all electrode sites under investigation (for more details see Appendix E2).

20 Just as for Experiment 2, the visual inspection of the time-frequency plots showed no systematic variations across conditions in higher frequency bands with regard to the applied measures.

Figure 4.5. Grand average ERPs and Gabor wavelet-based time-frequency difference plots in the delta band (1-3 Hz) for the non-related (Fig. 4.5.1) and related category conditions (Fig. 4.5.2) in comparison to the antonym control condition at electrode PZ (N=17). Figure 4.5.3 shows the difference between non-related and non-related category violations. The colour scale depicts the magnitude of the wavelet coefficient differences for EPow and WPow and the PLI value difference for PLI. Note that the violation conditions were subtracted from the antonym condition, thereby indicating relative decreases in activity for the former (cf. Exp. 1).

The global analyses for WPow and PLI again revealed significant main effects of TYPE at electrode PZ (WPow: F (2,32) = 20.01, p < .01; PLI: F (2,32) = 10.56, p < .01) and P4 (WPow: F (2,32) = 24.67, p < .01; PLI: F (2,32) = 10.34, p < .01). Post hoc pairwise comparisons showed that there were significant differences for related and non-related category violations in comparison to the antonym condition for WPow at PZ (NON: F (1,16)

= 23.77, p < .01; REL: F (1,16) = 24.70, p < .01) and P4 (NON: F (1,16) = 29.98, p < .01;

REL: F (1,16) = 31.40, p < .01), as well as for PLI at PZ (NON: F (1,16) = 14.74, p < .01;

REL: F (1,16) = 13.02, p < .01) and P4 (NON: F (1,16) = 10.57, p < .01; REL: F (1,16) = 16.28, p < .01). As for EPow (cf. Table 4.11), it is important to stress that there was no difference between both category violation conditions, neither with respect to WPow (F < 1) nor PLI (F < 1).²¹

21 All the results also hold for the remaining electrode sites (cf. Appendix E2).

Pz Delta

EPow Wpow PLI P4 Delta

EPow WPow PLI

TYPE ** ** ** TYPE ** ** **

NON x ANT ** ** ** NON x ANT ** ** **

REL x ANT ** ** ** REL x ANT ** ** **

NON x REL NON x REL

Table 4.11. Main effects of TYPE and pairwise comparisons for electrodes Pz and P4 with regard to the three measures applied for the delta frequency band (frequency bins: 1.0-3.0 Hz;

time window 100-500 ms). For post hoc single comparisons, all significance values are adjusted according to the modified Bonferroni procedure (Keppel, 1991) (** = < .01).

Figure 4.6 shows that, as for Experiment 2, there was also a stimulus-evoked increase in lower theta band activity (~3-5 Hz) between 300 and 600 ms. This increase was roughly in the same time range as (i.e. temporally overlapped with) as the delta EPow, WPow and PLI decrease (cf. Figure 4.5).

Figure 4.6. Grand average ERPs and Gabor wavelet-based time-frequency difference plots in the lower theta band (~3-5 Hz) for the non-related (Fig. 4.6.1) and related category violations (Fig. 4.6.2) in comparison to the antonym condition at electrode PZ (N=17). In terms of EPow- and PLI-differences, the non-related category violations (Fig. 4.6.1) but not the related category violations (Fig. 4.6.2) showed an EPow and PLI increase in comparison to the control condition. With regard to WPow, no power difference was observable at all; Note that the antonym condition was subtracted from each of the category violation conditions, thereby indicating relative increases in activity for the latter (cf. Exp. 1).

The statistical analysis indeed revealed a significant main effect of TYPE for EPow in the lower theta band (averaged frequency bins: 3.0-4.5 Hz) between 300 and 600 ms (PZ: F (2,32) = 4.22, p < .05; P4: F (2,32) = 6.07, p < .01). The resolution of the main effect showed that this effect was due to a significant increase in EPow for non-related category violations in comparison to antonyms (PZ: F (1,16) = 6.71, p < .05; P4: F (1,16) = 4.46, p < .05), whereas there was no significant difference for related category violations opposed to antonyms (F <

1).²² Additional analyses showed that there was no significant main effect of TYPE with regard to WPow (PZ: F (2,32) = 1.68, p < .3; P4: F < 1). However, the global analysis for the PLI revealed a significant main effect (PZ: F (2,32) = 6.05, p < .01; P4: F (2,32) = 8.97, p <

.01). Subsequent single comparisons unveiled that the increase in evoked power for non-related category violations compared to antonyms was due to an increase in phase-locking for the former (PZ: F (1,16) = 8.40, p < .01; P4: F (1,16) = 7.47 p < .05). Again, there was no significant difference for related category violations in comparison to antonyms (F < 1), but for related in comparison to non-related category violations (PZ: F (1,16) = 10.85, p < .01;

P4: F (1,16) = 33.24, p < .01). A schematic overview of the results is given in Table 4.12.²³

Pz

Lower Theta EPow Wpow PLI P4

Lower Theta EPow WPow PLI

TYPE * ** TYPE ** **

NON x ANT * ** NON x ANT * **

REL x ANT REL x ANT

NON x REL * ** NON x REL ** **

Table 4.12. Main effects of TYPE and pairwise comparisons for electrodes Pz and P4 with regard to the three measures applied for the lower theta frequency band (frequency bins: 3.0-4.5 Hz; time window 300-600 ms). For post hoc single comparisons, all significance values are adjusted according to the modified Bonferroni procedure (Keppel, 1991) (* = < .05, ** = < .01).

To compare the frequency-correlates of the late positivities observed here with the upper delta findings from Experiment 2, we also analysed the time-frequency matrices in the later time range. However, because the time window for the frontal positivity effect (450-600 ms) completely overlapped with the time window for the enhanced lower theta activity (300-600

22 Furthermore, there was a significant difference between related and non-related category violations (PZ: F (1,16) = 7.70, p < .01; P4: F (1,16) = 23.75, p < .01).

23 Again, the remaining electrodes showed the same pattern (cf. Appendix E2).

ms) and, furthermore, the lower theta frequency range (3.0-4.5 Hz) partially overlapped with the upper delta frequency range (2.0-3.5 Hz), we only analysed the late posterior positivity between 600 and 800 ms. Indeed, the statistical analysis revealed a significant main effect of TYPE in the upper delta band (2.0-3.5 Hz) for EPow (PZ: F (2,32) = 8.89, p < .001; P4: F (2,32) = 7.95, p < .005) and PLI (PZ: F (2,32) = 14.09, p < .001; P4: F (2,32) = 11.32, p <

.001). The resolution of the main effects showed that these effects were due to a highly significant enhancement of EPow (PZ: F (1,16) = 9.53, p < .01; P4: F (1,16) = 8.59, p < .01) and PLI (PZ: F (1,16) = 18.34, p < .005; P4: F (1,16) = 11.94, p < .005) for the non-related category words in comparison to antonyms. There was no significant effect for the related category words in comparison to antonyms (F < 1). Furthermore, an analysis of electrode sites Fz and Cz revealed that the enhanced upper delta EPow for non-related category words was more pronounced at posterior than at central or frontal electrode sites, analogous to the pattern observed for the late posterior positivity (cf. Table 4.13).

Fz

Upper delta

EPow Wpow PLI Cz

Upper delta

EPow WPow PLI

TYPE # # TYPE * **

NON x ANT * * NON x ANT * **

REL x ANT REL x ANT

NON x REL NON x REL * *

Pz

Upper delta

EPow Wpow PLI P4

Upper delta

EPow WPow PLI

TYPE ** ** TYPE ** **

NON x ANT ** ** NON x ANT ** **

REL x ANT REL x ANT

NON x REL ** ** NON x REL ** **

Table 4.13. Main effects of TYPE and pairwise comparisons for electrodes Fz, Cz, Pz and P4 with regard to the three measures applied for the upper delta frequency band (frequency bins: 2.0-3.5 Hz; time window 600-800 ms). For post hoc single comparisons, all significance values are adjusted according to the modified Bonferroni procedure (Keppel, 1991) (* = < .05, ** = < .01).

Summary

As for Experiment 2, we can distinguish three different EPow effects in comparison to the observed ERP pattern: (i) Both category violation conditions led to a decrease of stimulus-evoked activity in the delta band (1-3 Hz) in relation to the antonym condition. This decrease in EPow was due to a simultaneous decrease of whole power and phase locking. Most importantly, this decrease of delta activity was restricted to the time range in which the N400 effect was elicited. Furthermore, there was no significant difference between the two violation conditions in the delta band. (ii) In the lower theta band (3-4.5 Hz), solely the non-related category violation condition led to a significantly higher degree of stimulus-evoked activity in comparison to antonyms. This increase in EPow was due to an increase of phase-locking, whereas there was no effect with respect to WPow. With regard to the time range, this lower theta band EPow and PLI increase largely overlapped with the observed delta effect, and furthermore appeared almost simultaneously with the N400. However, as for Experiment 2, this lower theta EPow increase for non-related category violations extended at least until 700 ms post-onset of the critical word (i.e. until the time range of the late positivity), and furthermore showed a slight tendency to slow down in the later time range. Hence, the EPow modulation overlapped with the biphasic ERP pattern, thereby again rendering an unambiguous attribution to the N400 effect difficult. (iii) There was an increase in the upper delta band for the non-related category violation condition in comparison to antonyms in the P600 time range. This increase was clearly more pronounced at posterior electrode sites.

However, in comparison to Experiment 2, there was no significant upper delta effect for related category violations. Moreover, although the ERP pattern clearly showed a small but significant frontal positivity for both category violations, we couldn’t dissociate a respective correlate in the frequency domain due to a high degree of overlap with the preceding lower theta activity.

In sum, these findings strongly support the previous observations from Experiment 2 that, with regard to the N400 effect, there are two clearly distinguishable frequency bands which behaved rather differently with regard to their underlying frequency characteristics. On the one hand, there was a strong EPow modulation in the delta frequency band for antonyms in comparison to both category violation conditions. This EPow modulation was due to a parallel power and phase-locking increase. On the other hand, there was an increase of the

lower theta frequency for non-related category violations, which was due to an increase in phase-locking.