Revealing the neural networks that extract conceptual gestalts from continuously evolving or changing semantic contexts

Francesca M Branzi,Gina F Humphreys,Paul Hoffman,Matthew A Lambon Ralph

doi:10.1016/j.neuroimage.2020.116802

Francesca M Branzi, Gina F Humphreys + Show 2 more

Open Access

https://doi.org/10.1016/j.neuroimage.2020.116802

Copy DOI

Abstract

Reading a book, understanding the news reports or any other behaviour involving the processing of meaningful stimuli requires the semantic system to have two main features: being active during an extended period of time and flexibly adapting the internal representation according to the changing environment. Despite being key features of many everyday tasks, formation and updating of the semantic “gestalt” are still poorly understood. In this fMRI study we used naturalistic stimuli and task manipulations to identify the neural network that forms and updates conceptual gestalts during time-extended integration of meaningful stimuli. Univariate and multivariate techniques allowed us to draw a distinction between networks that are crucial for the formation of a semantic gestalt (meaning integration) and those that instead are important for linking incoming cues about the current context (e.g., time and space cues) into a schema representation. Specifically, we revealed that time-extended formation of the conceptual gestalt was reflected in the neuro-computations of the anterior temporal lobe accompanied by multi-demand areas and hippocampus, with a key role of brain structures in the right hemisphere. This “semantic gestalt network” was strongly recruited when an update of the current semantic representation was required during narrative processing. A distinct fronto-parietal network, instead, was recruited for context integration, independently from the meaning associations between words (semantic coherence). Finally, in contrast with accounts positing that the default mode network (DMN) may have a crucial role in semantic cognition, our findings revealed that DMN activity was sensitive to task difficulty, but not to semantic integration. The implications of these findings for neurocognitive models of semantic cognition and the literature on narrative processing are discussed.

Highlights

Successful time-extended semantic cognition relies on the ability of the semantic system to integrate information over time to build meaningful representations of the evolving world around us
Unlike prior investigations (e.g., Hasson et al, 2008; Lerner et al, 2011; Simony et al, 2016; Tylen et al, 2015), in this functional magnetic resonance imaging (fMRI) study we established which neural networks are primarily evoked by the formation and updating of semantic representations, and we distinguished them from those that support other non-semantic processes required during narrative reading
The semantic network We hypothesised that the building of the conceptual gestalt would be supported by two interactive neural systems, reflecting representational and control aspects of semantic cognition, respectively

Summary

Introduction

Successful time-extended semantic cognition (e.g., understanding the news reports on the radio or sequential paragraphs while reading a book) relies on the ability of the semantic system to integrate information over time to build meaningful representations of the evolving world around us. Instead, have measured multi-item semantic combinations without task manipulations that distinguish between brain structures involved in semantic integration (or formation of the semantic gestalt) from those involved in extra-semantic neuro-computations (e.g., working memory, attentional control, “schema formation” etc.: e.g., Hasson et al, 2008; Lerner et al, 2011; Simony et al, 2016; Tylen et al, 2015). Note that these latter studies have largely investigated the construct of “schema” or “situation model”, a representation that summarises the interaction among entities and the environment in a scene or event (e.g., Ranganath and Ritchey, 2012). The formation of a schema representation relies on a set of mnemonic processes that match incoming cues about the current context (for example, information about agents, space, time and social interactions) to these situation models

Methods

Results

Conclusion