Brain Rhythm Tracking May Share Predictive Neural Engine With Complex Human Syntax

In a newly published hypothesis and theoretical framework in Frontiers in Neuroscience, researchers Hangjun Chu and Tao Gong propose that musical beat perception and hierarchical syntactic structure building recruit a shared computational mechanism rooted in active, predictive neural entrainment. The authors argue that musical beat perception, defined as extracting and predicting a metrical pulse, and syntax, defined as assembling words into nested constituents while anticipating upcoming linguistic structure, both depend fundamentally on the brain’s ability to temporally anticipate unfolding events. Rather than viewing the empirical correlation between musical rhythm skills and grammatical ability as an incidental overlap, the framework synthesizes comparative neuroanatomy, infant development, and neuroimaging to suggest that a shared fronto-temporal predictive timing engine was functionally extended during human evolution to manage the nested demands of human language.

The theoretical model takes empirical motivation from developmental and behavioral research showing that rhythmic and grammatical abilities are strongly associated across childhood. Prior research in typically developing six-year-olds found that musical rhythm discrimination accounted for approximately half of the variance in expressive grammar after controlling for non-verbal ability and background factors, an association replicated in a subsequent investigation of 132 school-aged children. Furthermore, electroencephalography recordings from twenty-five six-year-olds engaging in passive listening to rhythmic tone sequences revealed that children exhibiting larger beat-evoked neural responses in the beta and gamma frequency bands scored higher on standardized tests of expressive grammar, particularly for structurally complex sentences. Experimental evidence from rhythmic priming studies has also demonstrated that brief exposure to regular musical rhythms can transiently improve grammaticality judgments in children with and without developmental language disorders, though the authors emphasize that such priming effects have not replicated uniformly across all languages, age groups, and experimental delays.

To account for these findings across phylogenetic and ontogenetic timescales, the authors propose a computationally nested three-layer architecture. Layer 1 is responsible for temporal encoding and regularity extraction, automatically extracting local statistical and temporal regularities directly from sensory input, such as syllable durations or tone transition probabilities. This layer represents an evolutionarily ancient and robust capacity present from birth, physiologically indexed by low-frequency neural tracking in the delta and theta bands that mirrors the acoustic structure of incoming stimuli. However, the authors emphasize that Layer 1 is necessary but not sufficient for syntax, as statistical tracking of local sensory transitions cannot autonomously generate nested structural hierarchies.

Operating on the representations provided by the lowest level, Layer 2 performs structural grouping and relational mapping. This intermediate computational stage groups sequential elements into discrete chunks and assigns abstract relational roles, such as establishing strong versus weak positions within a musical metrical beat or marking boundaries within continuous auditory streams. The authors propose that Layer 2 operates rapidly and automatically over short, familiar patterns and depends upon the dorsal auditory-motor pathway. The theoretical framework identifies three candidate anatomical implementations for Layer 2: higher-tier auditory cortex, where belt and parabelt fields could organize input prior to frontal engagement; inferior frontal cortex, following an architectural gradient analogous to visual processing stages; or dynamic, reciprocal functional interactions between auditory and inferior frontal regions connected by dorsal white-matter tracts.

The highest level of the architecture, Layer 3, carries out hierarchical prediction and internal model updating by recursively integrating Layer 2 chunks into multi-scale representations. This top-down predictive system concurrently forecasts upcoming structural elements across several hierarchical tiers, anticipating noun phrases or specific structural positions in spoken language just as it anticipates upcoming metrical downbeats in musical sequences. Functionally, Layer 3 is characterized by endogenous neural activity that cannot be explained by acoustic features alone, such as sensorimotor beta-band oscillations anticipating an unplayed beat or cortical tracking that persists when physical boundary markers are removed from speech. When incoming sensory input matches predictions, the internal model is reinforced; when mismatches occur, bottom-up prediction error signals prompt structural reanalysis and model revision.

From an evolutionary perspective, the authors draw on comparative primate neurobiology to argue that the hominin lineage functionally extended an ancient timing system rather than evolving entirely separate, de novo neural circuitry for syntactic computation. Non-human primates, including macaques and marmosets, demonstrate intact Layer 1 capabilities: they track sensory regularities, show neural sensitivity to isochrony, synchronize to metronomes with immediate feedback, and represent two-level auditory sequences across fronto-temporal structures. However, non-human primates show little evidence of spontaneous, beat-based metrical prediction or deeply nested sequence organization. Diffusion magnetic resonance imaging reveals that the human arcuate fasciculus possesses a substantially expanded projection into the temporal lobe and pronounced left-hemisphere lateralization compared to chimpanzees and macaques. Additionally, surface-based anatomical analyses indicate that human prefrontal gray matter volume is up to 1.9 times greater than that of macaques and 1.2 times greater than that of chimpanzees, while underlying prefrontal white matter exhibits differences of 2.4-fold and 1.7-fold, respectively. The authors hypothesize that this expanded top-down prefrontal drive over dorsal auditory-motor pathways enabled the ancestral timing apparatus to achieve the hierarchical predictive depth required for complex human syntax.

The developmental trajectory outlined in the paper treats ontogeny as logically independent from phylogeny, arguing that the coupling between beat prediction and grammar emerges alongside the protracted maturation of fronto-temporal connectivity. While infants possess rudimentary forms of all three layers at birth—including statistical learning, beat sensitivity demonstrated by neural responses to omitted downbeats, and early temporal anticipation—their predictions are initially restricted to short timescales and formulaic speech. Clinical dissociations observed in preterm children aged four to six illustrate that intact Layer 1 statistical learning can be preserved even when productive language lags behind term peers, indicating that lower-level regularity extraction cannot compensate for higher-level predictive bottlenecks. Because the arcuate fasciculus continues its myelination process into early adulthood, the authors hypothesize that the specific coupling between beat-based predictive entrainment and grammatical hierarchy emerges gradually between the preschool and early school years as white-matter pathways strengthen.

The authors formalize their hypothesis by detailing five falsifiable empirical predictions to test the limits of their model across diverse populations. Prediction 1 stipulates that non-human primates will continue to show intact Layer 1 tracking and two-level sequence prediction without endogenous beat anticipation, and that diffusion MRI measures of fronto-temporal connectivity should quantitatively predict the depth of predictive timing across species. Prediction 2 states that longitudinal developmental tracking will reveal an age-dependent emergence or strengthening of the correlation between beat-evoked predictive neural responses and syntactic processing that directly covaries with fronto-temporal white matter maturation. Prediction 3 posits that children with developmental language disorder will exhibit specific reductions in beat-evoked predictive responses to metrically structured rhythms compared to isochronous rhythms, reflecting an impairment in active predictive entrainment rather than a uniform sensory timing defect.

The final two predictions address plasticity and selective neuropsychological dissociations. Under Prediction 4, structured training protocols engaging multi-level metrical prediction are expected to generate significantly greater cross-domain improvements in syntactic processing than low-level isochronous tapping regimens. Finally, Prediction 5 establishes the critical falsification criterion using beat deafness, a condition where individuals fail to perceive musical meter despite intact pitch discrimination and normal synchronization to simple metronomes. The authors predict that beat-deaf individuals will exhibit subtle, measurable deficits in syntactic prediction, especially under working-memory load or during the processing of complex embedded clauses; conversely, finding completely normal syntactic prediction in individuals with confirmed deficits in active beat-based entrainment would fundamentally falsify the unified shared-resource hypothesis.

In evaluating the limitations of their synthesis, Chu and Gong emphasize that their proposal relies on evolutionary and developmental inferences that current data cannot definitively establish. Comparative differences in the arcuate fasciculus and behavioral timing between living primates provide an anatomical gradient, but whether structural connectivity directly caused the emergence of deep hierarchical prediction remains unproven. Similarly, because existing neural evidence linking beat processing to expressive grammar stems from a modest cross-sectional cohort of twenty-five children, the precise developmental age at which beat tracking couples to syntax remains an open empirical question. Furthermore, the authors note that their cortical framework must eventually be reconciled with subcortical timing networks, such as the basal ganglia and supplementary motor area, which are heavily engaged by rhythmic beats. By separating established empirical findings from theoretical inferences and outlining precise falsification criteria, the framework offers an actionable roadmap for investigating how the human brain harnesses rhythmic predictive timing to construct grammatical thought.

Subject of Research: Theoretical model linking musical beat perception to hierarchical syntactic processing via predictive neural entrainment

Article Title: From beat to syntax: evolutionary and developmental paths of a shared neural resource for hierarchical processing in music and language

Article References: Chu, H., & Gong, T. (2026). From beat to syntax: evolutionary and developmental paths of a shared neural resource for hierarchical processing in music and language. Frontiers in Neuroscience, 20, Article 1967182. https://doi.org/10.3389/fnins.2026.1967182

Image Credits: AI Generated

DOI: 10.3389/fnins.2026.1967182

Keywords: neural entrainment, beat perception, syntax, hierarchical processing, arcuate fasciculus, developmental neuroscience, predictive coding, auditory-motor pathway, beat, evolutionary, developmental, paths