Inconsistent and reconcilable differences across studies

The results from all four studies were consistent overall, especially with respect to the valence by arousal interaction and its direction;

nevertheless, some inconsistencies were also found. In particular, the prioritisation of emotionally valenced words compared to neutral words, reflected in RT as well as ERP measures, in both behavioural and ERP studies, was not confirmed by the RTs collected in the fMRI study, nor by the imaging results. However, these latter results might be somehow compatible with results from the previous two studies.

The higher activation observed for neutral words compared to negative in the inferior frontal gyrus suggests that neutral words require more effortful conscious processing for the lexical decision, as they are less salient than emotionally valenced words and cannot benefit from the recruitment of affective limbic regions. This is in line with the larger amplitude for neutral words found in the LPC, a component which reflects explicit processing of the stimulus prior to the response.

Nevertheless, it is not clear why the same difference was not found between neutral and positive words.

The RT inconsistency is less of a concern because behavioural data collected in the fMRI scanner are usually less reliable than data collected in a classical behavioural experiment. In fact, participants are in a lying position, cannot move their body or their head, and need to pay attention to this and many other possibly confounding factors while performing the task. Their response times are typically longer than in a more comfortable experimental setting (as was also the case in this research) and more variability is introduced.

6.2.1 Positive words are special. In the fMRI study a clear advantage of positive words over neutral and negative ones was not only observed in the RTs, but also in the BOLD responses, showing a modulation of arousal within positive words but not negative ones. This advantage could be explained by the fact that positive words have a more interconnected lexical and semantic network compared to negative, but also neutral words; therefore, they are easier to process (see Ashby, Isen, & Turken, 1999). These findings are compatible with an interaction of emotionality and hemisphere found in the LPC time-window, showing smaller amplitude for positive compared to neutral words, but also for positive compared to negative words, in the right hemisphere only. Also, the emotionality effect found in the early N1 component, showing larger amplitude for positive compared to neutral and negative words, might be considered to point in the same direction as the LPC effect, if we interpret larger amplitude in the N1 as early enhanced orientation of attention toward positive words.

If we also consider the findings regarding the familiarity (and self-referentiality) bias reported in both corpus and behavioural studies, the consistent processing advantage shown by positive material has two possible interpretations. First, positive words are naturally more salient than negative and neutral words given that healthy people show an automatic preference toward positive information (Fredrickson &

Branigan, 2005), which is also compatible with the idea of a more highly interconnected lexical and semantic network. Therefore, attention is automatically driven toward positive words even before matching of a lexical mental representation with the input stimulus, as reflected by the emotionality effect in the N1, but also by the activation of the insula, a region proposed as part of a “salience network” (Seeley et al., 2007).

This salience effect might have been enhanced in the ERP study by the specific material used: in fact the positive words were matched in arousal with negative words, but were slightly higher in absolute valence, i.e. “extremely valenced”. This was not the case for the material used in the fMRI study though, where positive and negative words were also matched for absolute valence.

An alternative interpretation, partly compatible with the first one, but more difficult to reconcile with the results from all studies, proposes that people show a bias toward positive material when asked to evaluate its familiarity or its possible reference to themselves (Citron et al., 2009;

Lewis et al., 2007). This might also be the case when they need to evaluate or judge verbal material with regards to different emotional or non-emotional properties (e.g. its lexicality). This interpretation would

reconcile the familiarity bias observed in the rating study with the emotionality by hemisphere interaction observed in the LPC, a component associated with stimulus evaluation; this is also in line with the general processing advantage observed in the imaging study, which cannot clearly establish whether the advantage occurred at a perceptual, attentional or evaluative processing stage. This interpretation is not compatible with the effect found in the N1 though, a component which does not index stimulus evaluation.

6.2.2. Subtle emotion effects. Generally, the psychophysiological results show subtle interactive effects of the two emotion dimensions, possibly due to the fact that the stimuli used were not intense enough, but only showed discrete degrees of differentiation across conditions.

This is partly due to the thorough manipulation and control of affective and lexico-semantic variables; for example, because negative words are usually more intense than positive stimuli (Citron et al., 2009; Lewis et al., 2007), matching them for arousal and for absolute valence led to the use of less intense negative words and less extremely valenced positive words (the latter one only in the fMRI study).

Subtle differences in the material led to subtle differences in brain activity, but the sensitivity of psychophysiological measures could have been improved by employing more trials in the ERP study, where only 25 stimuli per condition were employed in the valence by arousal design, as well as by the definition of regions of interest (ROIs) for the analysis of the fMRI data. In this study, the number of trials per condition was increased (N=35), but the analysis of the full brain might have very

likely hidden subtle emotional effects. By having a priori hypotheses on the regions responding to the experimental manipulation, the analyses can be tailored on specific areas and the likelihood of finding significant effects increases.

6.2.3. Other technical comparisons. Scalp distribution of the ERP effects and areas of activation found in the fMRI study cannot be compared; in fact the electro-cortical activity recorded from the posterior electrodes might originate from different cortical and sub-cortical regions, its distribution is only the output of brain activity detected by the electrodes. In order to identify the neural generator of this cortical activity, a source localisation technique can be used, which uses an algorithm to determine the possible source of electrical activity.

The neural generator could then be compared with the anatomical correlate found in the fMRI study.