This study investigates whether ACT-based impression formation equations can predict ACC activation in affectively incongruent yet semantically correct sentences describing social interactions. This investigation was motivated by a previous EEG study where we found an early effect of affective incongruency processing in the P2/N2 latency range [3]. We suggested this early effect to be a neural correlate of conflict detection which is related to ACC activation. This study's main goal was two-folded: a) to replicate previous findings from our EEG study and b) to test the hypothesis that the reported early effect of affective congruency violation is related to conflict processing as reflected by increased ACC activation. To ensure that expected findings tap into meaning making processes, we included attention questions and a control condition of hash signs (non-reading) to compare with brain activation in the experimental condition. As a manipulation check of affective congruency, we gathered likelihood ratings on the interactions illustrated in sentence stimuli. T-Test comparison confirmed well-known deflection-likelihood relations, i.e., events of greater affective deflection are perceived as less likely as compared to those of low affective deflection (12). Due to our highly controlled stimuli in the current design, this relation cannot be explained by other lexico-semantic variables known to influence expectancy related processes (cf. Schauenburg et al., 2019 ).
Taken together, behavioral and fMRI data strongly support the assumption that participants were indeed semantically engaged during sentence processing: answer accuracy to attention questions was above chance suggesting participants were attentively reading presented sentences and the results of whole level brain contrasts are in line with prevailing literature. Language comprehension and semantic processing are commonly associated with mainly left-lateralized activity in the lateral temporal lobe, including superior temporal gyrus (BA22), medial temporal gyrus (BA21), and temporal pole (BA38), the inferior frontal gyrus (BA44, 45), parietal lobe (BA39), posterior cingulate cortex (BA23) and prefrontal cortex (BA9), such as weaker activity in homologous areas of the right hemisphere [13–16]. We suggest that activity in caudate is linked to task demands depending on intention and attention-related processes necessary to keep presented linguistic representations active and suppress competing cognition [17]. Hippocampal activation has been shown to support flexible cognition related to online and incremental social and language processing crucial to successful behavior and creation of meaning of our world [18–20]. Therefore, we relate reported hippocampal activation to processes implicated in forming mental representations of presented social interactions, i.e., to bind and update prior knowledge of relevant concepts within actual, contextual and relational information.
Concerning effects of the relevant experimental manipulation on the neural level, we expected reading sentences with incongruent versus congruent affective meanings to correlate more strongly with hemodynamic responses in the ACC, a peculiar brain region monitoring, mediating and integrating cognitive and emotion-related processes facilitated by its specific connectivity to both limbic brain structures and the prefrontal cortex area [10]. Neural activity in ACC is correlated with error detection, conflict monitoring, expectancy-related processes, and affective experiences [6, 7, 21]. As hypothesized, ACC showed increased neural activity for affectively incongruent sentences. Thus, this study replicates evidence from our previous ERP results and provides a) neuroscientific evidence for the ACT-based mathematical model of impression formation of linguistic representations of social interactions, and b) suggests that violations of affective congruency are reflected by ACC activation even in the absence of explicit task demands, i.e., supporting the assumption of implicit, automatic processing of affective congruency [3].
In the current study, we used ACC activation especially as a dependent variable to assess the influence of affective incongruency on cognitive processing rather than investigating and adding new perspectives to the broad corpus of studies exploring basic mechanisms of ACC activation. However, here we want to give some considerations about how reported ACC activation to affective Deflection might be explained in terms of error signaling to prediction error. Recently, 22 [22] presented a computational model for mPFC and dlPFC. The Hierarchical Error Representation (HER) model is rooted in the tradition of Reinforcement Learning models which value in surprise as (prediction) error to be a key component in the formalization of association learning as the difference between maximum possible strength and current total strength [23]. By proposing that ACC is not sensible to the affective component of any stimulus, but to surprise (i.e., the discrepancy of the occurrence or non-occurrence of an expected or non-expected event), the HER model provides a considerable and unifying approach to explain ACC activation in various studies. Furthermore, the model assumes a hierarchical organization of frontal lobe processes where error signaling is driven by "bottom-up" and prediction error modeling by "top-down" processes. Within this approach, the reported finding of ACC activation can also be explained as an error signal of predicted affective congruency induced by sentence-final words. Crucially, the ACT-based concept of "affective Deflection" is somewhat similar to the concept of "error prediction" because it measures the degree to which an expected affective meaning is violated by a currently perceived affective meaning (which is further mirrored by the inverse correlation of affective Deflection and perceived likelihood of an event). Similar to the concept of prediction error, the degree of affective Deflection is an experienced-based correlate of the discrepancy between predicted probability and the actually observed social event.
Furthermore, ACT postulates that impressions of affectively incongruent events would initiate processes aiming to regain affective congruency [24]. Within the HER model, corresponding neural processes should be on "higher levels”, "top-down" modulating areas in the prefrontal cortex like dlPFC. A more detailed comparison of the HER model and ACT is beyond the scope of the current study. However, we assume that both models could complement each other concerning the computational modeling of precise and testable brain activation and connectivity to corresponding bottom-up and top-down processes throughout the frontal lobe.
Concerning the present results, we found increased activity only in the left ACC. Previous studies on conflict detection provide inconsistent findings concerning each hemisphere's differential contributions [25]. Crucially, 26 [26] found cognitive control to be localized in the same hemisphere as task execution: the left hemisphere was more involved in a letter-decision, whereas the right hemisphere was more involved in a visuospatial-decision task. Transferred to our study, this finding would explain activation of the left ACC due to the reading task associated with the left hemisphere. Nonetheless, 27 [27] found that differential hemispheric involvement was related to differences between cognitive processes of conflict detection versus error processing rather than to task-induced lateralized processing. 28 [28] reported increased left ACC activation due to conflicting stimuli in a cognitive flanker task, while the right ACC was more involved in an affective flanker task. However, going into more detail concerning the applied tasks and the salience of emotionality (implicit processing of affective congruency in a reading task vs. explicit categorization task of affective words), the comparison of reported findings of lateralized ACC involvement seems to be easier said than done. Future studies should further investigate differential lateralization of ACC activity in different task and stimulus settings to overcome these inconsistencies.
We consider the current study to be a fruitful and integrating approach of employing a social psychological model for an affective neuroscientific research question concerning the influence of affective language content on sentence processing. The applied concept of affective congruency as provided by ACT captures features of social information influencing linguistic representation of social events, which go beyond common variables known to influence semantic processing. However, some limitations need to be mentioned. First, we see that hash signs were probably not an ideal cognitive baseline in this design. They served as visually comparative stimuli, but what cognitive processes participants might be engaged in when perceiving hash signs is not strictly controlled. This also questions the comparability of processing demands in terms of task difficulty, see 14. Second, we presume that a combined EEG-fMRI study with the same group of participants would probably provide better-coupled information about the temporal evolvement of cognitive processes and their neural correlates. Third, individual differences in social cognition (e.g., empathy trait, theory of mind or mentalizing abilities), personality traits, or language processing competency could be investigated in future studies.
Our study is the first attempt to explore differential brain activity to manipulations of affective congruency as operationalized by ACT-based mathematical model of impression formation. Consistent with our hypothesis rooted in the reported findings of a previous EEG study, affective incongruency was related to increased ACC activation – even in the absence of an explicit processing task in semantically correct sentences. This study's observations emphasize the basic influence of affective language content during meaning making and are in line with the previous EEG study results.