Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural Oscillatory Dynamics of Hierarchical Syntax during Naturalistic Language Comprehension: An MEG Study
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Cheryl Gilford1, Balthasar Bickel1, Nina Kazanina2, Martin Meyer1; 1University of Zurich, 2University of Geneva
Syntactic structure building is fundamental to language comprehension, yet the role of neural oscillations in supporting this process during naturalistic listening remains poorly characterised. Electrophysiological studies have linked neural frequency bands to syntactic processing, though findings have varied across paradigms and analytic approaches. Moreover, previous studies have not used incrementally constructed syntactic representations during naturalistic speech comprehension, despite syntactic structure being continuously updated word-by-word during online comprehension (Meyer, 2018; Weissbart & Martin, 2024). To address this, we used a psycholinguistically plausible incremental parser to derive word-level syntactic features capturing hierarchical structure as linguistic input unfolds. Using a unified approach, we examine how five frequency bands, delta (0.5-4Hz), theta (4-8Hz), alpha (8-12Hz), beta (12-30Hz), and low-gamma (30-50Hz), relate to distinct stages of hierarchical syntactic structure building. Previously collected MEG data from 27 participants listening to four English narratives were analysed, with structural MRI data from 12 participants enabling source localization (Gwilliams et al., 2023). Syntactic structure was modelled using an incremental parser, generating word-level syntactic predictors (Roark et al., 2009). Five syntactic predictors were extracted, capturing distinct stages of hierarchical structure building, specifically incremental node count, subtree initiations, node closures, syntactic surprisal and syntactic depth of each word. Multivariate temporal response functions (mTRFs) were used to examine syntactic predictors encoding across frequency bands, with performance assessed using Pearson correlation and Bayesian hierarchical modelling. Additive and unique contribution analyses were conducted by comparing syntactic models against baseline control and reduced predictor models, allowing assessment of variance explained above controls and uniquely beyond other syntactic predictors. Results showed that all five frequency bands significantly encoded one or more parser-derived predictors above baseline, suggesting that syntactic information was represented across all bands with differing effects. Delta and beta activity showed additive and unique effects for subtree initiations and syntactic surprisal, respectively, suggesting functionally distinct roles in syntactic structure building. Delta activity can be linked to anticipating phrase-level constituents beyond acoustic tracking, and beta activity to updating or reanalysis of hierarchical representations in response to unexpected syntactic input. For some predictors, these bands did not improve model performance relative to baseline alone, but showed a unique effect once overlapping variance shared across syntactic predictors was accounted for. This suggests that their contributions may reflect more distributed aspects of syntactic processing that become apparent only when shared predictor variance is controlled for, indicating that these bands contribute more generally to overlapping aspects of syntactic structure building rather than being selectively associated with individual syntactic predictors (Martorell et al., 2023; Prystauka & Lewis, 2019). Source-level analyses revealed left-lateralised frontal, temporal, and parietal engagement. Syntactic surprisal effects in the beta and low-gamma bands showed stronger localisation to the left inferior frontal gyrus (IFG), suggesting greater involvement in syntactic reanalysis, whereas other syntactic predictors were associated more strongly with left temporal regions. Together, these findings suggest that different neural frequency bands support distinct and complementary computations involved in hierarchical syntactic processing during naturalistic speech comprehension, with time-resolved modelling revealing how these contributions unfold incrementally across continuous input.
Topic Areas: Syntax and Combinatorial Semantics, Speech Perception