Discussion

Hence, in the P600-time range, there was a stimulus-evoked activity enhancement at fronto-central electrode sites in the upper delta frequency range for both category violation conditions, which was due to a higher degree of phase-locking. Moreover, this upper delta EPow and PLI enhancement was more pronounced for non-related as compared to related category violations. For posterior electrode sites, however, only the non-related category condition led to a higher degree of stimulus-evoked upper delta activity. In addition this upper delta EPow effect was due to stronger phase-locking along with an increased whole power.

However, we already have stressed in Chapter 3.1.6 that the edges of the observed effects, despite the optimal time-frequency resolution of Gabor wavelets, might be subject to smearing effects, thereby rendering a precise demarcation of onsets and offsets in time and frequency difficult. Therefore, the distinction of an upper delta effect (in comparison to the lower theta frequency) is entirely tentative. It is based, firstly, on the visual evaluation of the time-frequency plots (i.e. the observation that there is a ‘bending’ of the theta frequency towards a lower frequency range in the time range of the late positivity effect). Secondly, it is based on the premise that a reflection of the observed P600-differences must also be present in EPow (and hence in WPow or PLI). Furthermore, the above objection clearly does not refer to the ‘centre of activation’ (activation peak) of an observed effect. That is, two clearly distinguishable activation centres are indicative of distinct effects in time and/or frequency.

However, the visual inspection of the time-frequency plots (cf. Appendix D1) for the electrodes Fz and, less pronounced, Cz clearly reveals distinct ‘activation peaks’ in time and frequency thereby supporting the tentative distinction between a lower theta and an upper delta effect.

violation conditions showed a pronounced biphasic N400-P600 pattern, it is obvious that, in this respect, the experimental manipulation was unsuccessful. As discussed in the final paragraph of Chapter 3, bi- or multiphasic ERP effects render the unambiguous attribution of frequency dynamics difficult, due to potential component overlap, concurrent superposition of frequency bands and smearing effects in the time-frequency domain due to the wavelet analysis. This is particularly problematic with respect to the analysis of low frequency dynamics, because of their inherently relatively poor temporal resolution.

Nevertheless, in comparison to the observed biphasic N400-P600 pattern, we could distinguish three different EPow effects in the frequency domain: (i) a pronounced increase in the delta band for antonyms in comparison to both category violation conditions in the N400 time range; (ii) an increase in the lower theta band for non-related category violations in comparison to antonyms and the related category violation condition in the N400 time range;

(iii) an increase in the upper delta band for both category violation conditions in comparison to antonyms in the P600 time range; this increase was more pronounced for the non-related than for the related category condition.

With regard to the pronounced N400 effect, we hypothesised that a large portion of this effect might be due to an enhanced positive ERP deflection for antonyms. That is, the N400 effect might not simply have been due to a reduced negativity for antonyms in comparison to both category violation conditions, but rather might have been the reflection of a nested positive component for antonyms. This hypothesis was strongly supported by the findings from the applied frequency measures. Under the assumption that the N400 effect is the result of a reduced negativity for antonyms in comparison to both category violations, the subtraction of the evoked power wavelet coefficients of the antonym condition (which served as a control condition) from the two category violation conditions should reveal a visibly enhanced activity for the latter. Furthermore, this enhanced activity should be graded, that is, more pronounced for non-related category violations than for related category violations (simply because the N400 effect showed this graduation). However, as was evident from Figures 4.2 and 4.3, there was no single correlate of the graded N400 effect observable in the frequency domain (in the time range of the N400 effect), which conformed to the above assumptions.¹⁵ The most important and eye-catching observation was that the EPow time-frequency

15 Of course, the assumption that the N400 effect is partly due to a reduced N400 component does not necessarily imply a single correlate in the frequency domain.

difference maps indicated a large positive difference in evoked delta power for antonyms in comparison to both category violation conditions (cf. Figure 4.2). In addition, this positive difference showed no graduation between related and non-related category violations. It was also evident that this difference was confined to the time range of the N400 effect. Even considering Gabor wavelet based smearing effects, which most likely blurred the precise temporal localisation of the on- and offset of the effects, this difference effect faded at approx.

500 ms post-onset of the critical item. Because the late positive portion of the biphasic N400-P600 effect didn’t start before 500 ms, it seems very unlikely that the difference effect in evoked delta activity was confounded with the latter effect. Therefore, this increase in evoked delta band power for antonyms can be unambiguously attributed to the N400 effect. Note that, in principle, the observed difference in evoked delta power must not necessarily reflect a delta power increase for antonyms in contrast to category violations. It also could be due to a more pronounced power decrease for category violations in comparison to antonyms. However, a comparison of the time course of delta power activation by means of time-frequency plots based on absolute EPow coefficient values revealed a clear event-related increase in delta power for antonyms. Hence, it is tempting to interpret this increase as a correlate of the positive shift in the N400 time range for antonyms and therefore as support for the hypothesis that the N400 effect is not a monolithic effect.

A second correlate of antonym processing was visible in the lower theta band. Here, the non-related category violation condition showed an increased evoked power in comparison to the antonym and the related category condition.¹⁶ Analogously to the graded N400 effect, this pattern allowed one to distinguish the two category violations from each other (i.e. to relate them with respect to their ‘gradedness’). However, in contrast to the delta band activity, the increase in the lower theta band can only be tentatively linked with the N400 effect. As the theta increase appeared between approximately 200 and 800 ms, it fully overllaped with the biphasic N400-P600 effect. Nevertheless, in the previous section, we already pointed to some arguments which suggest to us that an early and a late effect can be distinguished. First, between 500 and 800 ms post-onset of the critical stimulus, the frequency of the lower theta activity seemed to “slow down”. Additionally, visual inspection of the EPow time-frequency difference plots of electrode FZ revealed that there were clearly two distinct activation peaks for both category violation conditions in comparison to antonyms: an earlier increase in lower

16 There was only a marginal EPow increase in the lower theta band for related category violations (confined to electrode PZ and due to an increased PLI).

theta and a later increase in ‘upper delta’ (cf. Appendix D1).¹⁷ Hence, there was some evidence for the proposed distinction of an N400-related early lower theta increase and a P600-related late (upper) delta activity, despite the obvious smearing effects. Moreover, whereas the early EPow increase in lower theta was clearly due to an enhanced phase locking (without a concurrent increase in WPow), the picture for the late effect was more complex (but see above). We also proposed the distinction of a late frontal and a late posterior positivity. Although this distinction is tentatively supported by the topographical distribution of the observed upper delta effects, we do not discuss this issue in more detail in the present thesis. However, these findings will be subject to further investigations.

With regard to the primary aim of the present experiment, that is, to provide an exact and unequivocal characterisation of a semantic N400 effect, the lower theta EPow increase cannot unrestrictedly be attributed or related to the present N400 effect. This is the case although the evoked lower theta activity seems to be necessary to explain the graded effect between both category violation conditions.

In summary, the above observations constitute first evidence against the hypothesis that the present N400 effect is a monolithic effect (i.e. entirely due to a reduced N400 component for antonyms in comparison to both violation conditions). Instead, we hypothesised that the major proportion of the present N400 effect was due to an increased activity in the delta band for antonyms in comparison to the category violation conditions, thus reflecting an embedded positive component for antonyms. Moreover, the evoked delta power increase was due to an increase in whole power as well as in phase locking.¹⁸ Whereas an increased phase locking indicates a more consistent timing across trials, an increase in whole power can be taken as a reflection of an enhanced synchronisation of underlying neural populations. That is, increases in whole power might be a reflection of enhanced processing ‘effort’. However, further implication of the present results as well as a proposed functional interpretation will be

17 Note that this late frontal upper delta increase in EPow and PLI for category violations might be a reflection of the proposed distinction between a frontal and posterior late positivity (cf. Chapter 4.1.3; Hagoort et al., 1999;

Friederici, Hahne, & Saddy, 2002). However, because the focus of the investigation is on the N400, we will not further discuss the above findings with respect to late positivities in the context of the present thesis.

18 Unfortunately, if both WPow and PLI concurrently show activity changes, it is not possible (at least with the frequency measures applied here), to quantify the relative contributions of power or phase locking to EPow.

However, if we compare two or more conditions, the respective measures can show different degrees of activity relative to each other (cf. Experiment 1, Chapter 3).

discussed in the final section of the present chapter (together with and in the light of the findings from the subsequent experiments).

As already discussed above, the major problem of the present experimental manipulation was the somewhat unexpected appearance of a biphasic N400-P600 pattern (cf. Chapter 4.1.3), which rendered the desired frequency-analytical characterisation of the N400 effect, as well as its subsequent functional interpretation, difficult (at least with regard to the lower theta band).

Therefore, we have to consider the possible reasons for the elicitation of this late positivity. In general, the P600 is regarded as a correlate of syntactic processes (cf. Chapter 2; Friederici, 2002). However, the present antonymy sentence verification task undisputably involved semantic processing conflicts due to a semantic violation (i.e a wrong antonym relation), without any structural manipulations. Yet, late positivities have also been interpreted as a reflection of final evaluation processes with respect to the well-formedness of a sentence.

Hence, one could speculate that the embedding of antonym word pairs in a sentence context might have been responsible for the elicitation of a late positivity. Whereas a semantic violation like The honey was murdered is simply not plausible, i.e. is semantically odd (but conceivable in another possible world), the sentence The opposite of black is green is not only implausible but simply wrong (in all possible worlds). It not only contradicts our general world knowledge, but furthermore violates the whole sentence proposition (with regard to its inherent ‘logic’). In addition to the demands of the explicit antonymy verification task, it is very likely that the stereotype sentence context enhanced the tendency to build up a specific expectation with regard to the sentence-final words (the second antonym). This additional restricting factor might have enhanced the anticipatory aspect of the present task, thereby giving rise to an increased emphasis on evaluative or integrative processes. This in turn might have led to the elicitation of a late positivity. However, a further possibility might have been that the explicit task instruction to judge or verify the sentences with regard to their

‘underlying’ antonym relations led to the appearance of a late positivity. Nevertheless, we first conducted a further experiment in order to test whether the late positivity is still observable when antonym relations are presented out of sentence context. Therefore, in the subsequent experiment, we employed the same stimulus material as in the present experiment, but presented the antonym relations as word pairs.