Neural modelling of the semantic predictability gain under challenging listening conditions

Abstract

When speech intelligibility is reduced, listeners exploit constraints posed by semantic context to facilitate comprehension. The left angular gyrus (AG) has been argued to drive this semantic predictability gain. Taking a network perspective, we ask how the connectivity within language-specific and domain-general networks flexibly adapts to the predictability and intelligibility of speech. During continuous functional magnetic resonance imaging (fMRI), participants repeated sentences, which varied in semantic predictability of the final word and in acoustic intelligibility. At the neural level, highly predictable sentences led to stronger activation of left-hemispheric semantic regions including subregions of the AG (PGa, PGp) and posterior middle temporal gyrus when speech became more intelligible. The behavioural predictability gain of single participants mapped onto the same regions but was complemented by increased activity in frontal and medial regions. Effective connectivity from PGa to PGp increased for more intelligible sentences. In contrast, inhibitory influence from pre-supplementary motor area to left insula was strongest when predictability and intelligibility of sentences were either lowest or highest. This interactive effect was negatively correlated with the behavioural predictability gain. Together, these results suggest that successful comprehension in noisy listening conditions relies on an interplay of semantic regions and concurrent inhibition of cognitive control regions when semantic cues are available.

Keywords: angular gyrus; cingulo-opercular network; predictability gain; semantic network; speech-in-noise comprehension.

FIGURE 1

Design of the sentence repetition task. In each trial, participants listened to a sentence (black waveform) embedded in distracting background noise (grey waveform) whilst viewing a fixation cross on the screen. With the onset of the sentence's final keyword, a green traffic light prompted participants to orally repeat the whole sentence as accurately as possible within a 5 s recording period. The preceding sentence context was either predictive (green) or non‐predictive of the keyword (orange). Sentences varied orthogonally in intelligibility (ratio of speech intensity to SRT)

FIGURE 2

Behavioural results of the sentence repetition task. (a) Psychometric functions (coloured lines) were fitted to the proportion of correctly repeated keywords (coloured dots) across intelligibility levels for sentences with low (orange) and high predictability (green); grand average is displayed for illustration purposes. (b) The threshold of sentences with low predictability shifted towards more intelligible levels when compared to sentences with high predictability (p < .001; top). There was no evidence for a slope difference between sentences with low and high predictability (p = .122; bottom). Density plots illustrate the difference between the proportions correct of sentences with high vs. low predictability

FIGURE 3

Main effect of increasing speech intelligibility across the whole cortex. Top: Activation map thresholded at p < .05 (FWE‐corrected). Bottom: Average parameter estimates (percent signal change) for each experimental condition at peak voxels of left AG (MNI: x = −47, y = −67, z = 25), left IFG (MNI: x = −54, y = 33, z = 8) and right STG (MNI: x = 53, y = −12, z = 2). Error bars represent the standard error of the mean (SEM). AG, angular gyrus; IFG, inferior frontal gyrus; prec, precuneus; SFG, superior frontal gyrus; STG, superior temporal gyrus

FIGURE 4

Group‐level fMRI results illustrating brain regions that are sensitive to the interaction of intelligibility and predictability. (a) Parameter estimates illustrate the average interaction effect, extracted from peak voxels of DCM seed regions for the semantic network. (b) Brain regions showing a significant interaction; green = high > low predictable sentences; orange = low > high predictable sentences. The activation map is thresholded at p _{FWE‐corrected} < 0.05. (c) Parameter estimates illustrate the interaction effect, extracted from peak voxels of DCM seed regions for the cingulo‐opercular (CO) network. Error bars represent ±1 SEM. aIns, anterior insula; IFG, inferior frontal gyrus; MCC/PCC, middle and posterior cingulate cortex; PGa, anterior angular gyrus; PGp, posterior angular gyrus; pMTG, posterior middle temporal gyrus; pre‐SMA, pre‐supplementary motor area; SMG, supramarginal gyrus

FIGURE 5

Projection of activity onto the behavioural predictability gain across levels of intelligibility. (a) The condition‐specific proportion of behavioural performance explained by the interaction of intelligibility and predictability in each participant (middle line plot) was correlated with individual parameter estimates of single‐voxel BOLD activity (left and right bar plots; exemplary grand average parameter estimates for two peak voxels). Thin lines in the line plot represent single participants; fat lines represent grand average. Error bars represent ±1 SEM. (b) Group statistics revealed brain areas modulated by the single‐participant interaction effect of intelligibility and predictability on speech comprehension. p‐Values were FDR‐corrected within a mask of voxels responsive to the listening task (for an exhaustive list of clusters see Table 2). FP, frontal pole; PCG, paracingulate gyrus; PGp, posterior angular gyrus; pMTG, posterior middle temporal gyrus; PHG, parahippocampal gyrus; SCC, subcallosal cortex; SFG, superior frontal gyrus

FIGURE 6

DCM results. (a) Network architecture of the semantic (top) and cingulo‐opercular network (bottom); grey arrows with parameter estimates indicate significant intrinsic connections; black arrows indicate modulatory connections. Bar graphs show the modulation of connections by intelligibility (low, medium, high; binned) and predictability (low, high). The connection from PGp to PGa was significantly modulated by the main effect of intelligibility, the connection from pre‐SMA to left insula was sensitive to the interaction of predictability and intelligibility. Bar height represents grand average, error bars represent ±1 SEM. (b) Scatterplots (left) show the negative correlation of the condition‐wise predictability gain with the condition‐wise modulatory parameter estimates from pre‐SMA to left insula in two exemplary participants. Complementary, line plots (middle) illustrate patterns of behavioural performance (solid line) and modulation strength (dashed line) resolved for intelligibility (low, medium, high) and predictability (low = orange; high = green). Raincloud plot (right) shows the distribution of individual z‐transformed correlations across all participants; dots representing participants from scatter and line plots are highlighted in colour. n.s., not significant, *p < .05. l INS, left anterior insula; PGa, anterior angular gyrus; PGp, posterior angular gyrus; pMTG, posterior middle temporal gyrus; pre‐SMA, pre‐supplementary motor area; r INS, right anterior insula