Evidence for a deep, distributed and dynamic code for animacy in human ventral anterior temporal cortex

Abstract

How does the human brain encode semantic information about objects? This paper reconciles two seemingly contradictory views. The first proposes that local neural populations independently encode semantic features; the second, that semantic representations arise as a dynamic distributed code that changes radically with stimulus processing. Combining simulations with a well-known neural network model of semantic memory, multivariate pattern classification, and human electrocorticography, we find that both views are partially correct: information about the animacy of a depicted stimulus is distributed across ventral temporal cortex in a dynamic code possessing feature-like elements posteriorly but with elements that change rapidly and nonlinearly in anterior regions. This pattern is consistent with the view that anterior temporal lobes serve as a deep cross-modal ‘hub’ in an interactive semantic network, and more generally suggests that tertiary association cortices may adopt dynamic distributed codes difficult to detect with common brain imaging methods.

The left posterior angular gyrus is engaged by autobiographical recall not object-semantics, or event-semantics: Evidence from contrastive propositional speech production

“…In terms of the ventral ATL, there is strong evidence that the ATL is multi-modal semantic hub (86). For instance, the contribution of the ATL to speech production and comprehension is well-established in the neuropsychological literature whereby semantic dementia patients show significant deficits in both expressive and receptive semantic tasks (71) including impoverished content in propositional speech tasks (93,94) Furthermore, ATL engagement during semantic tasks has been consistently shown using fMRI and TMS in healthy participants and intracortical grid electrode studies in neurosurgical patients (10,(95)(96)(97)(98). Reliable ATL activation was less obvious in the meta-analysis (although still present at a reduced threshold).…”

A computational understanding of zoomorphic perception in the human brain

“…The human visual system is particularly sensitive for the distinction between inanimate and animate objects, the latter represented as a continuum (13)(14)(15)(16), and shows regions with a very strong selectivity for faces and bodies (9,31). However, these two types of selectivity are only partially captured by benchmark DNNs, possibly due to limitations in their training history.…”

“…In occipitotemporal cortex (VTC) there are distinct neural representations of categories such as faces (7,8), bodies (9,10), and animals (11,12). At the broader level, studies have found a hierarchically organized animacy continuum in VTC (13)(14)(15)(16), which has been related to perceptual and/or conceptual properties of animals (15)(16)(17)(18)(19).…”

“…However, there are many other properties of the human visual system and human visual experience that might more indirectly induce an Animal bias. In comparison to DNNs, human vision has been shown to be particularly sensitive to the superordinate distinction between animals and other objects (13)(14)(15)(16), to categories like faces and bodies (7)(8)(9)(10), to global shape (28), and to the natural statistics in the environment (29). None of these properties is captured fully in the benchmark DNN models in the literature.…”

Mapping lesion, structural disconnection, and functional disconnection to symptoms in semantic aphasia

“…SD patients show conceptual degradation associated with atrophy of the ventral anterior temporal lobes (vATL) and highly consistent semantic deficits across tasks that probe the same items (Bozeat et al, 2000; Mummery et al, 2000), in line with the view this site acts as a ‘semantic hub’ for heteromodal concepts (Lambon Ralph et al, 2017; Patterson et al, 2007). According to the hub-and-spoke model of semantic cognition, the vATL ‘hub’ works in concert with modality-specific ‘spokes’ in order to generate generalisable and coherent representations (Rogers et al, 2021). In contrast to those with SD, patients with SA have intact conceptual representations but an impaired ability to retrieve information in a flexible and context-appropriate manner, following left inferior frontal and/or temporoparietal stroke (Jefferies, 2013; Jefferies & Lambon Ralph, 2006) 1 .…”