Dynamic Predictions: Oscillatory Mechanisms Underlying Multisensory Sequence Processing

Abstract

Neural oscillations have been proposed to be involved in predictive processing by frequencyspecific modulation of either power or phase. While this is supported by substantial evidence on unimodal processing, only few studies are currently available that have addressed the role of oscillatory activity in multisensory predictions. In the present study, we have recorded MEG during a serial prediction task in which participants had to acquire stimulus sequences and to monitor whether subsequent probe items complied with the sequence. The sequences comprised combinations of visual and auditory stimuli, but on a given trial, only one of the modalities was task-relevant and the other had to be ignored. Task-related changes of power and coupling of neural oscillations were analyzed using a data-driven clustering approach. We observed prediction-related theta-band coupling in a network involving cingulate, premotor, prefrontal and superior temporal regions. Behavioral performance of the participants correlated with phase delays between remote regions in this network. Two additional clusters were detected where beta power were stronger when the previous or next stimuli were same as the current one in the sequence. These clusters interacted with the theta-band prediction network through crossfrequency coupling. Furthermore, we observed asymmetrical effects of attention to the visual and auditory modalities in a network distinguished by changes in alpha power, where alpha phases modulated response times. Overall, our results provide novel insights into the neural dynamics underlying multisensory predictions and suggest that oscillations in multiple frequency ranges as well as coupling within and across frequencies may be critical for sequence processing. Wang et al, Multisensory Sequence Prediction 3 KeywordsAttention, imaginary part of coherency, multisensory processing, phase-amplitude coupling.

Oscillatory and aperiodic neural activity jointly predict language learning

“…Here, fixed word orders contained linear order-based cues, which are analogous to English, while flexible word orders required animacy-based cues for interpretation. From this perspective, and in line with previous work on sequence processing (Crivelli-Decker et al, 2018; Kikuchi, et al, 2018; Wang et al, 2019), increased beta power likely reflected the propagation of top-down predictions during the learning of fixed word orders (Cross et al, 2018). In fixed sentences, the first noun is invariably the Actor, and as such, predictions are constrained to anticipating that the second noun will be the Undergoer, while also containing a verb-final construction.…”

Oscillatory and Aperiodic Neural Activity Jointly Predict Language Learning

“…Here, fixed word order sentences contained linear order-based cues, which are analogous to English, while flexible word orders required animacy-based cues for interpretation. From this perspective, and in line with previous work on sequence processing (Crivelli-Decker et al, 2018;Kikuchi, et al, 2018;Wang et al, 2019), beta synchronisation likely reflected the propagation of top-down predictions during the learning of fixed word order rules. Specifically, in fixed sentences, the first noun is invariably the Actor, and as such, predictions are constrained to anticipating that the second noun will be the Undergoer, while also containing a verb-final construction.…”

“…Previous research examining spectral dynamics during (artificial) language learning has revealed that increased alpha/beta power (i.e., synchronisation) predicts sensitivity to grammatical violations (Kepinska et al, 2017), while increased gamma phase coherence between frontal, temporal and parietal cortices is associated with successful learning (De Diego-Balaguer et al, 2011). An increase in theta synchronisation is also associated with poorer learning of auditory rules (De Diego-Balaguer et al, 2011).During visual grammar learning, theta synchronisation increases early in the task and declines after prolonged exposure (Kepinska et al, 2017).While these findings are inconsistent with images and word stimuli, studies using complex sequence processing have reported linear increases in beta power for predicable sequences, which coincide with increased theta activity in task-related cortical regions (Crivelli-Decker et al, 2018;Wang et al, 2019), and occipital alpha power (Wang et al, 2019). These data are also in line with the supposed role of active, top-down mechanisms in the processing of incoming sensory input, akin to a predictive-coding account of brain function (Friston, 2010;Kikuchi et al, 2018;Rao & Ballard, 1999).…”