Neural Correlates of Modal Displacement and Discourse-Updating under (Un)Certainty

Simplified illustration of main manipulations of experiments 1 and 2. Model of operations assumed to be present during the processing of factual (yellow) and modal (teal) statements (simplified from actual stimuli). Experiment 1 contrasts factual and modal statements in a factual discourse context, while experiment 2 varies whether the discourse context is factual, hypothetical, or presupposed. Updating of the discourse situation model (round) is expected to take place under certainty (in factual contexts with a factual update). Both modal (may) and conditional expressions (if superheroes wear masks) evoke hypothetical situations (cloud) involving modal displacement. Since the presupposed context marks information already known, we are not sure whether updating would take place.

Design and procedure experiment 1. A, Example stimuli set. Short narratives consisted of three parts. A context sentence biasing toward a rule-based or knowledge-based modal interpretation was followed by the target sentence containing one of the target verbs varying in force (possibility, necessity, or factual). The third continuation sentence was either congruent or incongruent with prior sentences. Details on controlled between-stimuli variation can be found in Extended Data Figure 3-1. B, Experimental design with number of items per condition in brackets (total = 240). The stimuli vary along two dimensions: Modal Base (rules, knowledge) and Force (possibility, necessity, factual). C, Continuation conditions. Half of the continuations are incongruent with the previous sentences. One third tap into modality and are congruent or incongruent with the modal base of the previous sentences. D, Trial structure with evoked MEG responses in femtotesla (fT) from one participant. A context sentence was displayed until participants pressed a button. After a fixation cross (300 ms), the target sentence was displayed word-by-word for 300 ms each followed by a 150-ms blank screen. The continuation sentence was displayed with a 600-ms delay, and participants indicated by button press whether this was congruent or incongruent with the prior story. Time windows for baseline correction (−2450 to −2250 ms) and statistiacal analysis (100−900 ms) are relative to the target verb (word6) onset.

Experimental design and procedure experiment 2. A, Example stimuli set and predictions. All stimuli were bi-clausal sentences of three different types: factual (p so q), conditional (if p → q), and presupposed (since p → q). These sentence types differed in whether they express information that is novel and certain (factual), novel and uncertain (conditional), or known and certain (presupposed). Each sentence contained either the factual verb do or the modal verbs may or might. Included are expected activation patterns for each verb per sentence type under processes of belief updating and modal displacement. We expect belief updating to take place in factual contexts but not in conditional contexts. For presupposed contexts, we had no clear predictions. Activity related to modal displacement is not expected to change across different sentential environments. B, Experimental design with number of items per condition displayed between brackets (total = 360). The stimuli vary among two dimensions: Sentence Type (factual, conditional, and presupposed) and Verb (may, might, do). C, Trial structure with evoked MEG responses in femtotesla (fT) from one participant. Procedure similar to experiment 1. Time windows for baseline correction (−3350 to −3200 ms) and statistical analysis (150–400 ms) are relative to the target verb (word8) onset.

Summary ROI results experiment 1 showing a main effect for factual over modal (possiblity and necessity) conditions in rIPS and TPJ, and an increase in activation for necessity in the rrACC. Results are collapsed for modal base (knowledge-based and rule-based modals grouped together). Boxplots display estimated brain activity within the time window of the identified temporal clusters, black dots indicate mean activity. ROIs are outlined on brain and shaded when containing identified clusters. Clusters significant after correction comparison across multiple ROIs indicated with asterisk and with grave accent when trending.

Time course of estimated average activity (dSPM) per modal force condition (factual, possiblity and necessity) for each ROI of experiment 1. Left hemisphere ROIs displayed on the left side, and right hemisphere on the right. Results collapsed for modal base (knowledge-based and rule-based modals grouped together). Detected clusters within time window 100–900 ms are highlighted and significance is indicated for the effect within the cluster (p_uncor) and when corrected for comparison across multiple regions (p_cor).

Identified spatiotemporal cluster of whole-brain analysis experiment 1. A, Time course estimated brain activity (dSPM) split by modal force condition (factual, possiblity and necessity) and identified cluster in gray. Boundaries of analysis window (100–900 ms) are indicated by dashed lines. B, FreeSurfer average brain shows spatial distribution of cluster, color shading indicating the sum of cluster-level F statistic (gained from cluster-based permutation test). C, Boxplots display estimated brain activity (factual > modal) within the identified time window of the spatiotemporal cluster, black dots indicate mean activity.

Time course estimated brain activity (dSPM) of reliable detected clusters from ROI analysis experiment 2. Both the lrACC and rvMPFC show an interaction between sentence type (factual, conditional, and presupposed) and verb (do, may, or might) with increased activation for do > may/might when embedded in factual sentences, and decreased activation for do < may/might in presupposed sentences. Boundaries of the analysis window (150–400 ms) are indicated by dashed lines, identified clusters displayed in gray. Boxplots display estimated brain activity within the time window of the identified temporal clusters, black dots indicate mean activity. ROIs are outlined on brain and shaded when containing identified clusters. Cluster effects are not significant after correction comparison across multiple ROIs. The effect in the lrACC was most prominent in the NY data while the effect in the rvMPFC was more prominent in the AD data (Extended Data Figure 8-1).