Pith. sign in

REVIEW 3 major objections 5 minor 37 references

Shadow of the (Hierarchical) Tree: Reconciling Symbolic and Predictive Components of the Neural Code for Syntax

T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper argues that the brain's neural code for syntax is a hybrid: low-frequency phase carries hierarchical phrase structure, high-frequency spiking carries linear statistics, and phase-amplitude coupling links the two.

desk verdict A coherent, wide-ranging Discussion piece that extends the author's ROSE architecture into a hybrid neurosymbolic model, but its headline LLM prediction is argued qualitatively, not derived from the math. read the letter →

arxiv 2412.01276 v1 pith:JKGNLG4C submitted 2024-12-02 cs.CL

Elliot Murphy This is my paper · ORCID
classification cs.CL
keywords syntaxneuralcodeROSEarchitecturephase-amplitudecouplinglargelanguagemodelsneurosymbolicpredictivecodinghierarchicalcomposition
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper tries to show that the brain's neural code for syntax is a genuine hybrid, not a competition between symbolic and statistical camps. It takes the ROSE neurocomputational architecture and assigns vertical, hierarchical phrase structure to low-frequency oscillatory phase codes, while horizontal, linear morphosyntax and lexical statistics are carried by high-frequency spiking and population codes. Phase-amplitude coupling is proposed as the concrete interface that lets symbolic tree structure shape statistical spiking. The most direct prediction is that large language models will match human brain activity for horizontal syntax but largely miss hierarchically compositional structure, which would redirect how such models are used in the cognitive neuroscience of language.

What carries the argument

The load-bearing mechanism is phase-amplitude coupling, mathematically expressed as the amplitude of high-frequency spiking activity being modulated by the phase of a low-frequency traveling wave: $A_{\mathrm{hi}}(x,\phi_L) = A_0(x) + A_m(x)\cos(\phi_L(t,x)-\phi_{\mathrm{pref}})$. In the paper's syntax-semantics model, a low-frequency oscillation encodes the syntactic category in its phase offset $\phi_c$, and the firing rate of each neuron becomes $P(S_i)[1+\alpha\cos(\phi_c(t)-\phi_{\mathrm{pref}})]$, so the population rate is the summed lexico-semantic statistics gated by a symbolic category. This makes phase-amplitude coupling the single interface where vertical symbolic code and horizontal statistical code meet, and it grounds the ROSE architecture's four levels (R, O, S, E) in one formal structure.

What would settle it

Record intracranial responses while listeners process sentences matched for word-by-word predictability but differing in phrase-structure configuration (e.g., identical word sequences with different attachment or headedness). If low-frequency phase measures such as inter-trial phase coherence do not systematically track the hierarchical condition, or if all distinguishing signal is instead carried by broadband high-frequency power, the ROSE-based hybrid model is falsified. A complementary falsifier: show that LLM activations predict neural responses in core language sites as strongly for nested hierarchical composition as for linear morphosyntax, contradicting the paper's central prediction.

Watch

Extended reading notes

Core claim

The paper's central claim is that the neural code for syntax is neurosymbolic: the same cortical circuitry uses a low-frequency phase code ($\delta$/ $\theta$) to represent vertical phrase structure and labeled syntactic categories, while spiking activity at high frequency encodes the statistics of lexical items, morphemes, and linear dependencies, with cross-frequency phase-amplitude coupling as the bridge between them. The author formalizes this with two models: one in which a low-frequency traveling wave modulates the amplitude of high-frequency spiking as a function of its phase, and one in which a phase-shifted oscillation encoding a syntactic category multiplicatively gates a population's firing rates for lexico-semantic features. On this account, symbolic rules are not stored redundantly but imposed dynamically through phase dynamics. The paper argues that LLMs succeed at the horizontal component because that is what surface statistics and attention can capture, while their documented failures in composition reflect the absence of the vertical symbolic channel.

Load-bearing premise

The argument depends on the assumption that low-frequency oscillatory phase in frontotemporal cortex genuinely carries hierarchical phrase-structure information, and that phase-amplitude coupling is how that structure is transmitted to spiking statistics; if low-frequency phase does not encode tree structure, the interface equations lose their grounding.

Editorial extensions

If this is right

  • LLM-brain alignment should be strong for horizontal morphosyntax and lexico-semantic statistics, but markedly weaker for hierarchically compositional structures.
  • Single-unit recordings will not find cells selective for individual syntactic features; feature-bundle coding plus spike-phase coupling handles syntax at the ensemble level.
  • $\delta$/ $\theta$-to-$\gamma$ phase-amplitude coupling in posterior temporal cortex should jointly track phrase-structure node closures and lexical surprisal, as MEG evidence reviewed in the paper suggests.
  • Experiments on the neural basis of syntax must control for statistical predictability before attributing effects to phrase structure, and vice versa.
  • The provided equations give an implementable recipe for injecting symbolic category information into a spiking neural regime, moving the symbolic-connectionist debate to concrete model space.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A direct corollary the paper leaves implicit: encoding models built from LLM embeddings should localize to sites and time windows tied to lexical/morphosyntactic statistics, while symbolic tree-feature models should win in windows tied to node closure and headedness—a testable double dissociation in the same dataset.
  • If the hybrid architecture is right, the same phase-amplitude interface may generalize beyond language to other hierarchical domains such as music and mathematical reasoning, where structure and statistics also coexist.
  • For neurosymbolic AI, the paper suggests a specific design principle: rather than bolting symbolic reasoning onto a network, let low-frequency phase variables gate the amplitude of high-frequency spiking activity, making symbol manipulation a dynamical state rather than a separate module.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. This Discussion article argues that the neural code for syntax is hybrid: lower levels of the author's ROSE architecture (Representation, Operation) encode statistical and predictive 'horizontal' information, while higher levels (Structure, Encoding) encode symbolic 'vertical' phrase structure, with phase-amplitude coupling as the interface. The paper reviews recent empirical work (Weissbart & Martin 2024; Zhao et al. 2024a, 2024b; Ten Oever et al. 2024; Dekydtspotter et al. 2024; Lakretz et al. 2024) as converging evidence for ROSE, discusses the syntax-phonology interface and predictive coding, and proposes a mathematical model in Sections 4.1–4.3 in which a low-frequency traveling wave modulates high-frequency spiking amplitude and a phase-shifted low-frequency oscillation gates population firing rates. The abstract and Section 4.5 state a central prediction: artificial language models will contribute to the cognitive neuroscience of horizontal morphosyntax, but much less so to hierarchically compositional structures. The paper concludes by advocating a neurosymbolic hybrid architecture for syntax.

Significance. If the empirical and conceptual claims hold, the paper would provide a valuable integrative framework for reconciling symbolic and connectionist approaches to syntax, and it offers a concrete falsifiable prediction about LLM-brain alignment that could guide future experiments. The review of recent intracranial, MEG, and EEG findings is broad and mostly accurate, and the paper makes a genuine effort to connect linguistic theory (MERGE, labeling) to neurobiological mechanisms. The mathematical models, however, are illustrative compatibility demonstrations rather than derivations or calibrated fits; the central asymmetry prediction about LLMs is not logically entailed by the equations, and the core premise that low-frequency phase encodes symbolic phrase structure is assumed from the author's own ROSE framework rather than independently established. The paper is therefore best read as a speculative synthesis with a partially ungrounded formal apparatus.

major comments (3)
  1. [Abstract; §4.5] The headline prediction—that LLMs will contribute to the cognitive neuroscience of horizontal morphosyntax but much less so to hierarchically compositional structures—is asserted in the abstract and Section 4.5 but is not derived from the mathematical model in Sections 4.1–4.3. Those equations describe phase-amplitude coupling between a low-frequency wave and high-frequency spiking activity, but no term distinguishes vertical from horizontal information, and no term represents LLM representations or their alignment with neural codes. The prediction rests on a qualitative analogy between transformer position/attention mechanisms and horizontal processes. Please either derive the asymmetry from the model with explicit formal assumptions or clearly label it as a separate empirical conjecture.
  2. [§4.1; §4.2] The premise that low-frequency oscillatory phase (delta/theta) in frontotemporal cortex encodes hierarchical phrase structure enters at the start of Section 4.1 from the ROSE framework and is not independently justified in this manuscript. The model is an unvalidated ansatz: the variables A_L, A_Hi, k, phi_L, phi0, alpha, and P(S_i) are free parameters with no constraints from data, and no fitting, simulation, or comparison with empirical measurements is provided. If low-frequency phase does not carry symbolic syntactic information, the interface equations lose their semantic grounding. Please add explicit discussion of this assumption's empirical status and what evidence would disconfirm it.
  3. [§4.3] The 'integrated model for ROSE' stacks four separate equations into a vector F(R,t) and calls this a coupled evolution of the ROSE state, but the four components are not actually coupled in the equations: there is no term linking dR/dt, O_k(t), the tree representation, and the Kuramoto phase dynamics. The sentence 'The state representation of ROSE at time t would evolve under coupled differential equations' is therefore misleading. Either specify the coupling terms explicitly or describe the model as four independent component models.
minor comments (5)
  1. [Throughout] There are numerous typographical errors and citation inconsistencies, including 'theoretial' (§1), 'horiztonal' (§1.3), 'parients' (§2.2), 'freqeuncy' (§4.1), 'Valzquez' (§3.2), 'Mizadeh' (§4.5), 'Antisymetry' (reference), 'Dobashy' (reference), 'Sefllers' (reference), and 'perfetly' (§2). A careful proofread is needed.
  2. [§4.1] The equation for H(t,x) is typeset with a leading ';' and other formatting artifacts that make the mathematics hard to read; the same applies to the summation in §4.3. The authors should provide clean-LaTeX versions of all equations.
  3. [References] Several cited works do not appear in the reference list (e.g., 'Shi 2024' is referred to in §3.2 but only 'Sho, E.R. (2024)' appears, and 'Lang, Y. (2021)' and 'Levelt et al. (1999)' are listed but not cited in the text). Please reconcile all citations.
  4. [§4.4] The Hopf algebra section is briefly sketched and does not clarify how the proposed oscillatory product M(x,y)=A_x A_y e^{i(phi_x+phi_y)} relates to the comultiplication Delta(x)=x⊕z or to neural data. This section would benefit from a concise statement of what empirical observation would distinguish this formalism from a purely descriptive notation.
  5. [§5] The concluding Baudelaire epigraph and the Nobel Prize reference, while evocative, do not add scientific content and could be removed for conciseness.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: ROSE is an explicit premise, the LLM prediction is an interpretive synthesis of cited evidence, and the Section 4 equations are constructive postulates rather than derived predictions.

full rationale

The paper does not reduce its central claims to its inputs by construction. ROSE is explicitly treated as the starting architecture ('Building on a recent neurocomputational architecture for syntax (ROSE)'), not as a conclusion derived in this paper; the author states that the reader is referred to the original position paper for details. The key LLM asymmetry claim ('artificial language models will contribute to the cognitive neuroscience of horizontal morphosyntax, but much less so to hierarchically compositional structures') is presented as one prediction of the hybrid framing, but it is supported by cited behavioral and neural evidence about LLM strengths and limitations (e.g., Linzen & Baroni 2021; Marvin & Linzen 2018; Dentella et al. 2024; Huang et al. 2024), not obtained from the equations in Sections 4.1-4.3. This is an interpretive claim with an evidence gap, not a circular derivation. The mathematical model in Sections 4.1-4.3 is a constructive compatibility model: it assigns low-frequency phase to symbolic labels ('symbolic features... can be represented by the spatial gradient (kx) of the wave') and high-frequency amplitude/rate codes to statistical features. Because these mappings are explicit assumptions of the model, restating them as 'demonstrating' an interface is unpacking a postulate rather than secretly assuming the conclusion. The author is transparent that these are ways ROSE could be formalized, and Section 5 concedes the proposal is 'a model, like any other, with clear limitations.' Self-citations such as Murphy (2020b, 2024) identify the source of ROSE's assumptions, whereas key empirical anchors come from independent studies (Weissbart & Martin 2024; Ten Oever et al. 2024; Lakretz et al. 2024; Dekydtspotter et al. 2024). No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the author's prior work, and no known empirical pattern is merely relabeled. The absence of a formal derivation of the LLM prediction is a methodological weakness, but it is not circularity.

Assumptions & free parameters 7 free parameters · 5 assumptions · 3 invented entities

The framework depends heavily on the author's own ROSE architecture and on Chomskyan assumptions about Merge and labeling. The mathematical model introduces a large number of unconstrained parameters (wave amplitudes, frequencies, phases, coupling strengths, projection matrices) without fitting them to data. The central mechanistic claim, that phase-amplitude coupling links a symbolic phase code to statistical spiking activity, is an assumption rather than a derived result.

free parameters (7)
  • A_L (low-frequency wave amplitude)
    Scale of the traveling wave in Eq. (1); chosen by hand and not fitted to data.
  • f_L, k, phi_L (frequency, wavenumber, phase offset)
    Parameters of the traveling wave; free parameters of the model.
  • A_Hi, f_Hi, phi_Hi (high-frequency component parameters)
    Parameters of spiking activity; chosen to represent statistical morpheme codes.
  • A0i(x), A_mod_i(x), phi_psi (baseline amplitude, modulation strength, preferred phase)
    Define the phase-amplitude coupling function; all free.
  • alpha (coupling strength)
    Strength of phase modulation of spike rates; the paper notes it could be learned via a Hebbian rule but does not specify values.
  • P(S_i) (semantic feature weights)
    Statistical lexico-semantic features; conceptually derived from embeddings or frequency statistics but not operationalized.
  • W, b (projection matrix and bias)
    Parameters for the dimensionality reduction step in Section 4.3; unspecified.
assumptions (5)
  • ad hoc to paper The ROSE architecture (Representation, Operation, Structure, Encoding) correctly describes how the brain represents hierarchical syntax.
    ROSE is introduced in the author's prior work and is the foundational framework throughout the paper.
  • domain assumption Merge and labeling are the core computational operations of syntax.
    Adopted from Chomskyan generative grammar; stated in Section 2.1.
  • ad hoc to paper Phase-amplitude coupling can carry symbolic syntactic information between oscillatory phase codes and spiking statistical codes.
    This is the key mechanistic assumption underlying the mathematical models in Sections 4.1 and 4.2.
  • domain assumption Syntax is partitioned into vertical hierarchical structure and horizontal linear/statistical information.
    The vertical/horizontal distinction is used throughout and is supported by some cited empirical work, but it is a contested framing.
  • standard math Low-frequency oscillations modulate high-frequency spiking via sinusoidal phase-amplitude coupling.
    The sinusoidal modulation is a standard model of phase-amplitude coupling, but its applicability to symbolic syntax is assumed.
invented entities (3)
  • ROSE architecture levels (R, O, S, E) independent evidence
    purpose: To explain how atomic features, operations, structural inferences, and workspaces implement syntactic computations in the brain.
    Introduced in Murphy (2024); the present paper cites MEG/EEG and single-unit studies as converging evidence, but the architecture itself is author-postulated and not directly measured.
  • Phase-encoded syntactic category (phi_c)
    purpose: To map syntactic labels (NP, VP) onto a low-frequency phase signal.
    A modeling construct without a direct independent measurement; it is a placeholder for whatever neural signal carries category information.
  • Ephaptic coupling as interface for statistical inference independent evidence
    purpose: To mediate probabilistic information flow in the proposed hybrid model.
    Ephaptic coupling is a documented physical phenomenon, but its specific role in linguistic statistical inference is speculative and not tested.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Shadow of the (Hierarchical) Tree: Reconciling Symbolic and Predictive Components of the Neural Code for Syntax." pith.science (2026). https://pith.science/paper/JKGNLG4C

@misc{pith2026241201276,
  author       = {Pith},
  title        = {Pith review of: Shadow of the (Hierarchical) Tree: Reconciling Symbolic and Predictive Components of the Neural Code for Syntax},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JKGNLG4C}},
  note         = {Machine review of arXiv:2412.01276}
}
read the original abstract

Natural language syntax can serve as a major test for how to integrate two infamously distinct frameworks: symbolic representations and connectionist neural networks. Building on a recent neurocomputational architecture for syntax (ROSE), I discuss the prospects of reconciling the neural code for hierarchical 'vertical' syntax with linear and predictive 'horizontal' processes via a hybrid neurosymbolic model. I argue that the former can be accounted for via the higher levels of ROSE in terms of vertical phrase structure representations, while the latter can explain horizontal forms of linguistic information via the tuning of the lower levels to statistical and perceptual inferences. One prediction of this is that artificial language models will contribute to the cognitive neuroscience of horizontal morphosyntax, but much less so to hierarchically compositional structures. I claim that this perspective helps resolve many current tensions in the literature. Options for integrating these two neural codes are discussed, with particular emphasis on how predictive coding mechanisms can serve as interfaces between symbolic oscillatory phase codes and population codes for the statistics of linearized aspects of syntax. Lastly, I provide a neurosymbolic mathematical model for how to inject symbolic representations into a neural regime encoding lexico-semantic statistical features.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 33 canonical work pages

  1. [1]

    els of linguistic structure. Importantly, the vertical-leading model I am discussing here is fully compliant with typical assumptions in syntax-phonology interface research, according to which certain types of syntactic information is accessible to phonology, but syntax is strictly phonology-free (Frota & Vigário 2018). This forms the core of Pullum and Z...

    work page 1988

  2. [4]

    A Hybrid Neural Code for Syntax Many previous models of natural language syntax have been either wholly symbolic (Chomsky 2013, 2023; Chomsky et al. 2019,

    work page 2013

  3. [5]

    [a] middle ground must be found

    Conclusion Happy is he who can with a vigorous wing Propel towards the luminous and serene realms; He whose thoughts, like larks, Free, in the morning take flight, — Hover over life, and understand with ease The language of flowers and silent things! Charles Baudelaire, ‘Élévation’, Les Fleurs du mal (1861) A number of avenues – experimental and theoretic...

    work page 2022

  4. [7]

    Murphy, E. (2024). ROSE: A neurocomputational architecture for syntax. Journal of Neurolinguistics 70: 101180. Murphy, E. (2025). The nature of language and the structure of reality. In Benítez-Burraco, A., López, I.F., Férnandez-Pérez, M., & Ivanova, O. (eds.), Biolinguistics at the Cutting Edge: Promises, Achievements, and Challenges. Berlin: De Gruyter...

    work page 2024

  5. [9]

    (at the E level). 25 (7) Spectral information like phase and oscillatory cycles could offer a brain-intrinsic coordinate system by which the output of neurocognitive patterns can be interpreted (van Bree 2024). (8) Top-down predictions about higher-order structure imposed by ROSE could be implemented via firing rate functions, non-linear postsynaptic resp...

    work page 2024

  6. [13]

    2022; Piantadosi & Gallistel 2024)

    for lexico-semantic features being accessed by S complies with demands of minimal description (Dehaene et al. 2022; Piantadosi & Gallistel 2024). These lexical feature bundles at R can be represented at O (𝑂.) by γ-coordinated (spike-phase coupled) synchronously firing neural populations (where 𝑟#(𝑡) = spike train of neuron 𝑖; 𝑂.(𝑡) = firing rate of the p...

    work page 2022

  7. [14]

    2024, but see Slaats et al

    to encode hierarchical structures at S – or, for bonus psycholinguistic plausibility, left-corner recurrent neural network grammars (Sugimoto et al. 2024, but see Slaats et al. 2024; I suspect that different parsing models will have their performance impacted by whether we apply them to low-frequency components or high-frequency 𝛾; e.g., bottom-up parsing...

    work page 2020

  8. [15]

    In summary, current connectionist models of syntax typically exhibit a strong bias for syntactically simple linear-distance units (Lakretz et al

    – in stark contrast to many other cognitive systems and perception-action cycles. In summary, current connectionist models of syntax typically exhibit a strong bias for syntactically simple linear-distance units (Lakretz et al. 2019). This might account for why they perform well on horizontal linguistic representations but less well on vertical, hierarchi...

    work page 2019

Show all 37 references
  1. [16]

    little evidence of sophisticated phrase composition

    will be grounded in low-frequency phase codes that flexibly adapt their tuning to lower-order sensorimotor features via spike-phase coupling. I refer the reader to Dentella et al. (2024) for a comprehensive assessment of the grammatical and semantic competence of LLMs, but her...

    2024

  2. [19]

    with frameworks that highlight the role of language in inference generation, cognitive/generative model updating, consolidation of experience, endogenous planning and monitoring, aiding directed attention, reflection of personal experience and values, and compositional semanti...

    2011 arXiv

  3. [66]

    Zioga, I., Weissbart, H., Lewis, A.G., Haegens, S., & Martin, A.E. (2022). Naturalistic spoken language comprehension is supported by alpha and beta oscillations. Journal of Neuroscience 43: 3718–3732

    2022

  4. [134]

    Cecchetto, C., & Donati, C. (2015). (Re)labeling. Cambridge, MA: MIT Press. Chaves, R.P. (2020). What don’t RNN language models learn about filler-gap dependencies. Proceedings of the Society for Computation in Linguistics 3: 20–30. Chen, H., Ye, T., Zhang, M., & Ding, N. (202...

    2015 doi

  5. [139]

    Chowdhury, S., & Zamparelli, R. (2018). RNN simulations of grammaticality judgments on long-distance dependencies. Proceedings of the 27th International Conference on Computational Linguistics 133–144. Collins, A., & Loftus, E. (1975). A spreading activation theory of semantic...

    2018 doi

  6. [154]

    Murphy, E., Hoshi, K., & Benítez-Burraco, A. (2022a). Subcortical syntax: Reconsidering the neural dynamics of language. Journal of Neurolinguistics 62: 101062. 49 Murphy, E., & Leivada, E. (2022). A model for learning strings is not a model of language. Proceedings of the Nat...

    2022

  7. [287]

    Yeaton, J.D. (2024). The neurobiology of sentence production: a narrative review and meta-analysis. PsyArXiv: https://doi.org/10.31234/osf.io/xku24. Yu, L., & Ettinger, A. (2020). Assessing phrasal representation and composition in transformers. Proceedings of the 2020 Confere...

    2024 doi

  8. [469]

    Narita, H. (2014). Endocentric Structuring of Projection-free Syntax. Amsterdam: John Benjamins. Neeleman, A., & van de Koot, J. (2006). On syntactic and phonological representations. Lingua 116: 1524–1552. Nelson, M.J., El Karoui, I., Giber, K., Yang, X., Cohen, L., Koopman, ...

    2014

  9. [705]

    Murphy, E., Woolnough, O., Rollo, P.S., Roccaforte, Z., Segaert, K., Hagoort, P., & Tandon, N. (2022b). Minimal phrase composition revealed by intracranial recordings. Journal of Neuroscience 42(15): 3216–3227. Narain, D., Remington, E.D., Zeeuw, C.I.D., & Jazayeri, M. (2018)....

    2022

  10. [723]

    Dehaene, S., Al Roumi, F., Lakretz, Y., Planton, S., & Sablé-Meyer, M. (2022). Symbols and mental programs: a hypothesis about human singularity. Trends in Cognitive Sciences 26(9): 751–766. Demberg, V., Keller, F., & Koller, A. (2013). Incremental, predictive parsing with psy...

    2022

  11. [724]

    Keshev, M., Cartner, M., Meltzer-Asscher, A., & Dillon, B. (2024). A working memory model of sentence processing as binding morphemes to syntactic positions. In L. K. Samuelson, S.L. Frank, M. Toneva, A. Mackey, & E. Hazeltine (eds.), Proceedings of the 46th Annual Conference ...

    2024 arXiv

  12. [1423]

    tokens of processes with physiological bounds that render them into functional types

    calls “tokens of processes with physiological bounds that render them into functional types”. I have previously discussed the potential role of ephaptic coupling (Murphy 2015b) in an emerging neural infrastructure for language, and with the introduction of ROSE we can now retu...

    2010

  13. [1515]

    Murphy, E. (2015c). Reference, phases and individuation: topics at the labeling-interpretive interface. Opticon1826 17(5): 1–13. Murphy, E. (2018). Interfaces (travelling oscillations) + recursion (delta-theta code) = language. In E. Luef, & M. Manuela (eds.), The Talking Spec...

    2015

  14. [2013]

    (where 𝑟#(𝑡) = spike rate of the 𝑖-th neuron; 𝜆#(𝑡) = baseline firing rate that reflects semantic statistical properties; 𝜂#(𝑡) = noise term, e.g., Gaussian or white noise): 𝑟#(𝑡)=𝜆#(𝑡)+𝜂#(𝑡) Relevant lexico-semantic statistics can be distributed across the population (where 𝑃...

    2022

  15. [2014]

    # = frequencies of high-frequency components; 𝐴

    levels. An interesting topic concerns when and where ‘modular’ or ‘transform’ strategies for Bayesian inference in the brain are harnessed for syntactic and semantic prediction (Sohn & Narain 2021). More generally, the core architectural component of ROSE, phase-amplitude coup...

    2021

  16. [2018]

    2022; Ganguli & Simoncelli

    and population (Dubreuil et al. 2022; Ganguli & Simoncelli

    2022

  17. [2019]

    across levels of linguistic information, with phase-based predictions being tested against amplitude-based sensorimotor transformations. Cross-frequency coupling (n:m), as implemented in ROSE, serves to segregate cortical computation into distinct but interfacing processing re...

    2024

  18. [2020]

    in artificial models. Studies of human reading that explicitly probe for syntactic disambiguation effects show that probabilistic measures like surprisal are not sufficient to explain task difficulty (Huang et al. 2024). That the brain uses probabilistic information to infer s...

    2024

  19. [2021]

    inevitably generates surprisal-related and entropy-related measures for lexical feature selection, opening up questions about psycholinguistic factors, which we now turn to. 3.3. Symbolic and Predictive Components: Psycholinguistic Factors In everyday cases, the use of statist...

    2021

  20. [2023]

    we must develop systems that combine the expressiveness and procedural versatility of symbolic systems with the fuzziness and adaptiveness of connectionist representations

    or wholly probabilistic (Kwiatkowski et al. 2012). Some recent models have tried to incorporate both of these 23 (Marcus 2001; Martin 2020; Murphy et al. 2024a), and it is in this spirit that I have approached this topic here (see also Pessoa 2022). In this section, my treatme...

    2001

  21. [2024]

    As reviewed extensively elsewhere (Murphy 2015b, 2018, 2020b, 2024), this accords with the consensus that lower frequencies help segregate faster rhythms (Northoff & Huang 2017)

    this permits a connection between lower-order statistical inferences encoded in γ to 20 be mapped to a symbolic phase-coupled lower frequency code. As reviewed extensively elsewhere (Murphy 2015b, 2018, 2020b, 2024), this accords with the consensus that lower frequencies help ...

    2018

  22. [2101]

    Murphy, E. (2020b). The Oscillatory Nature of Language. Cambridge: Cambridge University Press. Murphy, E. (2021). Linguistic Representation and Processing of Copredication. PhD thesis, University College London. Murphy, E. (2023). The citadel itself: defending semantic interna...

    2020

  23. [2446]

    Callaway, E. (2024). Chemistry Nobel goes to developers of AlphaFold AI that predicts protein structures. Nature 634: 525–526. Caucheteux, C., Gramfort, A., & King, J.R. (2022). Deep language algorithms predict semantic comprehension from brain activity. Scientific Reports 12:...

    2024

  24. [2768]

    Goto, N., & Ishii, T. (2024). Seeking an optimal design of Search and Merge: its consequences and challenges. The Linguistic Review 41(1): 119–152. 43 Grabot, L., Kononowicz, T.W., la Tour, T.P., Gramfort, A., Doyère, V., & van Wassenhove, V. (2019). The strength of alpha-beta...

    2024

  25. [4850]

    Berlin: Springer. 9–17. 50 Northoff, G., & Huang, Z. (2017). How do the brain’s time and space mediate consciousness and its different dimensions? Temporo-spatial theory of consciousness (TTC). Neuroscience & Biobehavioral Reviews 80: 630–645. Orhan, P., Boubenec, Y., & King, ...

    2017 arXiv

  26. [5240]

    Forseth, K.J., Kadipasaoglu, C.M., Conner, C.R., Hickok, G., Knight, R.T., & Tandon, N. (2023). A lexical semantic hub for heteromodal naming in middle fusiform gyrus. Brain 141(7): 2112–2126. Forseth, K.J., Pitkow, X., Fischer-Baum, S., & Tandon, N. (2021). What the brain doe...

    2023 doi

  27. [6606]

    Gwilliams, L., Linzen, T., Poeppel, D., & Marantz, A. (2018). In spoken word recognition, the future predicts the past. Journal of Neuroscience 38: 7585–7599. Gwilliams, L., Marantz, A., Poeppel, D., & King, J.R. (2024). Hierarchical dynamic coding coordinates speech comprehen...

    2018 doi

  28. [8850]

    Woolnough, O., Donos, C., Murphy, E., Rollo, P.S., Roccaforte, Z.J., Dehaene, S., & Nitin, T. (2023). Spatiotemporally distributed frontotemporal networks for sentence reading. Proceedings of the National Academy of Sciences of the United States of America 120(17): e2300252120...

    2023

  29. [9878]

    the electric field enslaves neurons, not the other way around

    put it, “the electric field enslaves neurons, not the other way around”. A similar logic motivates ROSE, where lower frequencies act to scaffold local neural dynamics. Low-frequency oscillations create rhythmic fluctuations in the extracellular electric field, which can phase-...

    2021

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.