Research study: "When inefficient speech-motor control affects speech comprehension: atypical electrophysiological correlates of language prediction in stuttering" (2021)

Research study: "When inefficient speech-motor control affects speech comprehension: atypical electrophysiological correlates of language prediction in stuttering" (2021) Research: [https://www.biorxiv.org/content/10.1101/2021.10.28.466231v1.full](https://www.biorxiv.org/content/10.1101/2021.10.28.466231v1.full) PDF document: [https://www.biorxiv.org/content/10.1101/2021.10.28.466231v1.full.pdf](https://www.biorxiv.org/content/10.1101/2021.10.28.466231v1.full.pdf) ## Abstract It is well attested that people predict forthcoming information during language comprehension. The literature presents different proposals on how this ability could be implemented. Here, we tested the hypothesis according to which language production mechanisms have a role in such predictive processing. To this aim, we studied two electroencephalographic correlates of predictability during speech comprehension ‒ pretarget alpha‒beta (8-30 Hz) power decrease and the post-target N400 event-related potential (ERP) effect, ‒ in a population with impaired speech-motor control, i.e., adults who stutter (AWS), compared to typically fluent adults (TFA). Participants listened to sentences that could either constrain towards a target word or not, allowing or not to make predictions. We analyzed time-frequency modulations in a silent interval preceding the target and ERPs at the presentation of the target. Results showed that, compared to TFA, AWS display: i) a widespread and bilateral reduced power decrease in posterior temporal and parietal regions, and a power increase in anterior regions, especially in the left hemisphere (high *vs*. low constraining) and ii) a reduced N400 effect (non-predictable *vs*. predictable). The results suggest a reduced efficiency in generating predictions in AWS with respect to TFA. Additionally, the magnitude of the N400 effect in AWS is correlated with alpha power change in the right pre-motor and supplementary motor cortex, a key node in the dysfunctional network in stuttering. Overall, the results support the idea that processes and neural structures prominently devoted to speech planning and execution support prediction during language comprehension. ## Introduction Research on human language has traditionally treated production and comprehension as separate systems, with minimal interaction. This view has been supported by findings in classic aphasiology and language acquisition, showing asymmetries in these abilities in brain-lesioned patients and children. However, recent research suggests that these asymmetries do not necessarily indicate separate systems. Instead, the neural substrates for comprehension and production largely overlap, implying shared processes and representations. ## Focus on Prediction This study investigates prediction, a crucial aspect of integrated language approaches. Language comprehension involves both bottom-up and top-down processes, where comprehenders actively predict forthcoming information. This prediction is thought to be linked to production processes, particularly through covert simulations using motor-to-sensory forward models, which are used to monitor one's own speech and predict incoming speech during comprehension. ## Relevance of Stuttering People who stutter (PWS) provide a suitable population to test the hypothesis that speech production mechanisms contribute to prediction during comprehension. Developmental stuttering (DS) involves disrupted speech flow due to deficits in planning, timing, and executing speech-motor sequences. PWS show abnormalities in various brain regions associated with speech production and in white matter structures of speech-motor pathways. ## Neurocomputational Models The DIVA/GODIVA and HSFC models highlight the importance of sensorimotor integration in speech production, where motor plans predict sensory consequences. DS may result from impaired feedforward processing or inefficient motor-to-sensory predictions, leading to an overreliance on feedback control systems. ## Electrophysiological Evidence Electrophysiological techniques, such as EEG and MEG, show that alpha (8-12 Hz) and beta (13-30 Hz) power decreases are associated with speech planning and execution. In PWS, these frequency bands are abnormally modulated, indicating inefficient motor-to-sensory transformations and poor coordination of speech production regions. ## Hypothesis and Experimental Design Given the impaired speech-motor control in DS, it is hypothesized that PWS will exhibit atypical prediction processes during comprehension. The study focuses on two electrophysiological correlates of predictability: the pre-target alpha-beta power decrease and the post-target N400 event-related potential (ERP) effect. ## Pre-Target Power Decrease In typical populations, alpha and beta power decreases are observed before predictable words during comprehension, suggesting shared processes for predicting and producing language. Previous studies have shown that these power decreases are correlated with predictability in frontal brain regions involved in speech production, which are also implicated in DS. ## N400 Effect The N400 ERP effect is a robust marker of predictive processing, elicited by predictable versus non-predictable words. Previous studies have found a reduced N400 effect in PWS in response to semantic violations, indicating impaired semantic integration. However, the N400 response to non-anomalous predictable versus non-predictable words in PWS has not been investigated. ## Methodology The experiment involved adults who stutter (AWS) and typically fluent adults (TFA) performing a sentence comprehension task. The predictability of the final word was manipulated by the preceding sentential context, with sentence frames being either highly (HC) or low constraining (LC). Oscillatory activity was analyzed in the silent interval before the target word, and ERPs were computed at the onset of the target word, focusing on the N400 time-window (300-500 ms post-target). ## Expected Outcomes If DS affects predictive efficiency during comprehension, AWS are expected to show a reduced alpha-beta power decrease before the target word and a reduced N400 effect compared to TFA. This would support the hypothesis that speech production mechanisms contribute to predictive processes during language comprehension. ## Discussion This study explored electrophysiological correlates of predictability—pre-target alpha-beta power decrease and post-target N400 effect—between adults who stutter (AWS) and typically fluent adults (TFA). The aim was to determine if speech-motor control deficits in AWS affect prediction during spoken language comprehension. Findings revealed that AWS exhibit less consistent alpha-beta power decreases before predictable words, mainly in posterior areas, and a significantly reduced N400 effect compared to TFA. These results indicate inefficient prediction processes in AWS. ## Alpha-Beta Power Decrease * **Findings in AWS**: The alpha-beta power decrease was less pronounced and primarily restricted to left posterior areas, with a notable absence in anterior regions. * **Findings in TFA**: Both anterior and posterior regions showed power decreases. * **Group Comparison**: Differences in power modulation suggest AWS and TFA use different prediction mechanisms. AWS showed reduced power decreases in posterior areas and power increases in frontal areas, though the latter might not be statistically significant. * **Functional Interpretation**: Alpha-beta power decreases are linked to cortical engagement and information richness, suggesting AWS have less precise pre-activated information. Power increases in AWS may indicate inhibitory control to prevent incorrect activations, aligning with known deficits in inhibitory processing in AWS. ## N400 Effect * **Findings in AWS**: Both groups showed an N400 effect, but it was attenuated in AWS. The difference was driven by responses to predictable words, suggesting less efficient top-down facilitation of bottom-up processing in AWS. * **Functional Interpretation**: The reduced N400 effect indicates less efficient auditory prediction, likely due to malfunctioning speech-motor systems. ## Correlations Between Power Modulations and N400 Effect * **AWS**: Positive correlations were found between alpha power in right supplementary and premotor areas and N400 amplitude, suggesting less cortical engagement is associated with a reduced N400 effect. These regions are crucial in the speech-motor network, indicating a role in coordinating sensorimotor information for prediction. * **TFA**: A power decrease in the beta range in the left inferior frontal cortex correlated with a stronger N400 effect, highlighting this region's involvement in prediction. ## Implications and Theoretical Models * **Speech Production and Prediction**: The results support the DIVA/GODIVA model, emphasizing the integration of speech production and self-monitoring in prediction processes. * **AWS Deficits**: The findings suggest that suboptimal feedforward performance in AWS's sensorimotor systems leads to inefficient planning and prediction, affecting proactive processing during speech comprehension. ## Conclusions The study provides evidence that speech-motor control deficits in AWS impact their ability to predict during spoken language comprehension. This inefficiency is reflected in altered alpha-beta power modulations and a reduced N400 effect, pointing to broader implications for understanding the neural underpinnings of stuttering and its effects on language processing. We studied the electrophysiological correlates of prediction during auditory language comprehension in adults who stutter. We found evidence for less efficient prediction in this population. This study adds novel evidence in support of the involvement of the neural infrastructure devoted to speech-motor control in prediction during speech comprehension. The study ultimately stresses two major points: i) a more integrated investigation of language comprehension and production is desirable, and we advocate for a view in which the two modalities are seen as task sets drawing from at least partially shared resources, rather than two separate systems (see e.g., [**McQueen and Meyer, 2019**](https://www.biorxiv.org/content/10.1101/2021.10.28.466231v1.full#ref-58)), and ii) the investigation of on-line speech/language comprehension in DS, in addition to speech/language production, should be further pursued, in order to attain a better understanding of the neural architecture of language in healthy and/or pathological conditions. Specifically, regarding DS, we suggest that the role of executive functions and inhibitory control should be further investigated and taken as potential targets for better speech therapy interventions, extending treatment techniques using cognitive and linguistic approaches in addition to approaches mainly focused on “classical” fluency-shaping techniques. Future research should investigate which representations and processes are implicated in different tasks (and how), in order to better specify the language architecture and language use. This, in turn, can help in the development of more effective therapeutic approaches to developmental and acquired disorders of both language comprehension and production.