Elliot Murphy
@elliot-murphy
Postdoc, Dept of Neurosurgery, UTHealth Compositionality in neural and artificial systems Website:
Lastly, we assessed the relationship between low frequency coding and gamma power. Gamma and low frequency ramping effects were mostly spatially dissociable at the single electrode level, with only 20% showing substantial evidence of a gamma ramping response also showing low frequency ramping.
To quantify interactions across this network showing alpha frequency modulation we used Granger causality between each ROI pair, revealing remarkable consistency across participants and consistent dominant flow direction from lateral temporal cortices to the lateral and medial frontal lobe.
We plot the temporal progression of frequency and power shifts across 6 core ROIs that emerged from these analyses.
We applied sbLME models contrasting sentences and word lists to quantify the interaction effect between word position and whether the words were in a structured sentence or unstructured wordlist (300-500ms after each word onset), and measured differences at the end of the sentence.
To evaluate interactions between LN and DMN, we extracted instantaneous frequency, power, and phase (4-30 Hz) from intracranial electrodes in participants who read sentences and wordlists.
There is no contradiction here. Finding mean gamma signatures spread across cortex for node structure when you regress out word frequency means that those effects are very likely not driven by word frequency, but they *can* also be driven by many other things beyond word frequency! (ROSE screenshot)
The authors rightly note that my 2024 ROSE paper leaned away from attributing causal-explanatory power to mean gamma power for constituent/structural identity. I was also a co-author on the Woolnough et al study, which *did* in fact find gamma effect of syntactic node structure (see Suppl Figs).
New paper out today on syntactic constituency structure tracking in EEG! www.sciencedirect.com/science/arti...
Our results provide a systems-level account for how the human brain transforms word forms into conceptual meaning that is invariant of subjective judgment. We offer a summary model to go beyond traditional hub-and-spoke architectures.
Cortical stimulation of ventrotemporal cortex and inferior frontal cortex disrupted the ability of patients to make concreteness judgements, indicating some causal role of these regions in the task.
Words occupying the middle range of the concreteness scale (e.g., ‘profit’, which can be subjectively rated as either con/ab) activated similar regions, but high-frequency signatures were not modulated by the participants’ semantic decisions about these words.
Inter cortical communications, measured by high-frequency partial direct coherence, revealed bidirectional frontal and ventral plus lateral temporal interactions.
We chart the temporal profile of regional sensitivity to concreteness, showing widespread cortical engagement.
Interestingly, the pMTG (a region previously implicated in our work in phrase structure composition and sentence processing) showed greater engagement for abstract words over concrete words, suggesting its involvement in language-specific representations AND computations.
Concrete concepts showed greater high-frequency activation across a frontal and ventrotemporal network, while greater activation for abstract words was found in lateral posterior middle temporal cortex.
Models of conceptual processing have been limited in spatiotemporal resolution, and uncertainty remains about the causal role of specific regions in concept representation. We utilized intracranial recordings in human neurosurgical patients with epilepsy during a concreteness judgement paradigm.
New book chapter, 'The quo vadis of the relationship between language and large language models', with Vittoria Dentella and @evelinaleivada.bsky.social! In an excellent new volume on language and AI with contributions from Charles Yang, David Lobina, Jordan Kodner et al vernonpress.com/book/2434
ChatGPT thinks I am not able to hold a pen in my hand without full stability.
On your other point, we have previously addressed this issue more than a decade ago by showing that the activations in the brains of individuals with epilepsy are not dissimilar compared to healthy volunteers across semantic classes (nouns and verbs) (Conner et al. Cerebral Cortex 2014) (5/10)
With “thinking” on it only gets it if I directly show the photo, not from the logical semantic description — so more “seeing” and less “thinking”. Mine also fails with this counting to 200 example. youtube.com/shorts/7_2zk...
Published today - my review of Greg Hickok's excellent new book 'Wired for Words'! @gregoryhickok.bsky.social bioling.psychopen.eu/index.php/bi...
We are hiring ✨ Epilepsy and Functional Neurosurgery Fellowship 🧠 ⚡️
We take as a case study cross-linguistic influence (CLI) and attendant theories of functional inhibition/competition, and present these as being driven by specific oscillatory failure modes during L2 sentence planning.
The concept of raising a glass of sherry for one of your concepts 🧠 🍸
