Elliot Murphy

The commentary articles responding to Murphy (2025) are thoughtful, generous, and appropriately demanding. Each of them offers creative approaches to assessing neurocomputational architectures for ...

Futrell and Mahowald argue that neural networks have learned non-trivial aspects of language. We argue that these systems have not in fact demonstrated “mastery” of syntax, marshalling recent evidence, and that they further obscure explanatory insights with respect to topics in the cognitive neuroscience of language.

Neurobiological 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 ( n = 19) during a concreteness judgement paradigm of single word reading. 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. Intercortical communications, measured by high-frequency partial direct coherence, revealed bidirectional frontal and ventral plus lateral temporal interactions. Words occupying the middle range of the concreteness scale (e.g., “profit”) activated similar regions, but high-frequency signatures were not modulated by the participants’ semantic decisions about these words. Cortical stimulation of ventrotemporal cortex and inferior frontal cortex disrupted the ability to make concreteness judgements. These results suggest that semantic information is encoded via a causally directed system of bidirectional cortical cascades: early visual-linguistic integration in ventrotemporal cortex initiates directed information flow to frontal hubs, and later processing shows reciprocal flow back to ventral and lateral temporal regions integrating distinct conceptual features, with these convergence zones differing based on semantic type. Our results provide a systems-level account for how the human brain transforms word forms into grounded conceptual meaning that is invariant of subjective judgment.

In the modern language sciences, the core computational operation of syntax, 'Merge', is defined as an operation that combines two linguistic units (e.g., 'brown', 'cat') to form a categorized constituent structure ('brown cat', a Noun Phrase). This structure can be further combined with additional linguistic units based on this categorial information, respecting non-associativity such that abstract grouping is preserved. Some linguists have embraced the view that Merge is an elementary, indivisible operation that emerged in a single evolutionary step. From a neurocognitive standpoint, different mental objects constructed by Merge may be supported by distinct tiers of processing demands: (1) simple command constructions (e.g., "eat apples"); (2) the merging of adjectives and nouns ("red boat"); and (3) the merging of nouns with spatial prepositions ("laptop behind the sofa"). Here, we systematically investigate participants' comprehension of sentences with increasing levels of syntactic complexity. Clustering analyses revealed behavioral evidence for three distinct structural types, which we discuss as potentially emerging at different developmental stages and subject to selective impairment. While a Merge-based syntax may have emerged suddenly in evolutionary time, responsible for the structured symbolic turn our species took, different processing tiers seem to underwrite the comprehension of various types of Merge-based objects.

A core component of a successful artificial general intelligence would be the rapid creation and manipulation of grounded compositional abstractions and the demonstration of expertise in the family of recursive hierarchical syntactic objects necessary for the creative use of human language. We evaluated the recently released o3 model (OpenAI; o3-mini-high) and discovered that while it succeeds on some basic linguistic tests relying on linear, surface statistics (e.g., the Strawberry Test), it fails to generalize basic phrase structure rules; it fails with comparative sentences involving semantically illegal cardinality comparisons ('Escher sentences'); its fails to correctly rate and explain acceptability dynamics; and it fails to distinguish between instructions to generate unacceptable semantic vs. unacceptable syntactic outputs. When tasked with generating simple violations of grammatical rules, it is seemingly incapable of representing multiple parses to evaluate against various possible semantic interpretations. In stark contrast to many recent claims that artificial language models are on the verge of replacing the field of linguistics, our results suggest not only that deep learning is hitting a wall with respect to compositionality (Marcus 2022), but that it is hitting [a [stubbornly [resilient wall]]] that cannot readily be surmounted to reach human-like compositional reasoning simply through more compute.

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.