The relationship between communicated language and intended meaning is often probabilistic and sensitive to context. Numerous strategies attempt to estimate such a mapping, often leveraging recursive Bayesian models of communication. In parallel, LLMs have been increasingly applied to semantic parsing applications, tasked with inferring logical representations from natural language. While existing LLM explorations have been largely restricted to literal language use, in this work, we evaluate the capacity of LLMs to infer the meanings of pragmatic utterances. Specifically, we explore the case of threshold estimation on the gradable adjective “strong”, contextually conditioned on a strength prior, then extended to composition with qualification, negation, polarity inversion, and class comparison. We find that LLMs can derive context-grounded, human-like distributions over the interpretations of several complex pragmatic utterances, yet struggle composing with negation. These results inform the inferential capacity of statistical language models, and their use in pragmatic and semantic parsing applications.

Recent developments in the biology of malignant gliomas have demonstrated that glioma cells interact with neurons through both paracrine signaling and electrochemical synapses. Glioma-neuron interactions consequently modulate the excitability of local neuronal circuits, and it is unclear the extent to which glioma-infiltrated cortex can meaningfully participate in neural computations. For example, gliomas may result in a local disorganization of activity that impedes the transient synchronization of neural oscillations. Alternatively, glioma-infiltrated cortex may retain the ability to engage in synchronized activity in a manner similar to normal-appearing cortex but exhibit other altered spatiotemporal patterns of activity with subsequent impact on cognitive processing. Here, we use subdural electrocorticography to sample both normal-appearing and glioma-infiltrated cortex during speech. We find that glioma-infiltrated cortex engages in synchronous activity during task performance in a manner similar to normal-appearing cortex but recruits a diffuse spatial network. On a temporal scale, we show that signals from glioma-infiltrated cortex have decreased entropy, which may affect its ability to encode information during nuanced tasks such as production of monosyllabic versus polysyllabic words. Furthermore, we show that temporal decoding strategies for distinguishing monosyllabic from polysyllabic words were feasible for signals arising from normal-appearing cortex but not from glioma-infiltrated cortex. These findings inform our understanding of cognitive processing in chronic disease states and have implications for neuromodulation and prosthetics in patients with malignant gliomas.

Two analytic traditions characterize fMRI language research. One relies on averaging activations across individuals. This approach has limitations: because of inter-individual variability in the locations of language areas, any given voxel/vertex in a common brain space is part of the language network in some individuals but in others, may belong to a distinct network. An alternative approach relies on identifying language areas in each individual using a functional 'localizer'. Because of its greater sensitivity, functional resolution, and interpretability, functional localization is gaining popularity, but it is not always feasible, and cannot be applied retroactively to past studies. To bridge these disjoint approaches, we created a probabilistic functional atlas using fMRI data for an extensively validated language localizer in 806 individuals. This atlas enables estimating the probability that any given location in a common space belongs to the language network, and thus can help interpret group-level activation peaks and lesion locations, or select voxels/electrodes for analysis. More meaningful comparisons of findings across studies should increase robustness and replicability in language research.

Gliomas synaptically integrate into neural circuits^1,2. Previous research has demonstrated bidirectional interactions between neurons and glioma cells, with neuronal activity driving glioma growth^1-4 and gliomas increasing neuronal excitability^2,5-8. Here we sought to determine how glioma-induced neuronal changes influence neural circuits underlying cognition and whether these interactions influence patient survival. Using intracranial brain recordings during lexical retrieval language tasks in awake humans together with site-specific tumour tissue biopsies and cell biology experiments, we find that gliomas remodel functional neural circuitry such that task-relevant neural responses activate tumour-infiltrated cortex well beyond the cortical regions that are normally recruited in the healthy brain. Site-directed biopsies from regions within the tumour that exhibit high functional connectivity between the tumour and the rest of the brain are enriched for a glioblastoma subpopulation that exhibits a distinct synaptogenic and neuronotrophic phenotype. Tumour cells from functionally connected regions secrete the synaptogenic factor thrombospondin-1, which contributes to the differential neuron-glioma interactions observed in functionally connected tumour regions compared with tumour regions with less functional connectivity. Pharmacological inhibition of thrombospondin-1 using the FDA-approved drug gabapentin decreases glioblastoma proliferation. The degree of functional connectivity between glioblastoma and the normal brain negatively affects both patient survival and performance in language tasks. These data demonstrate that high-grade gliomas functionally remodel neural circuits in the human brain, which both promotes tumour progression and impairs cognition.