Graded constraints in semantic cognition: How do we retrieve knowledge in a flexible way?

Informations projet

FLEXSEM

N° de convention de subvention: 771863

DOI

10.3030/771863

Projet clôturé

Date de signature de la CE 21 Mars 2018

Date de début 1 Avril 2018

Date de fin 30 Septembre 2024

Financé au titre de

• EXCELLENT SCIENCE - European Research Council (ERC)

Coût total

€ 1 999 860,00

Contribution de l’UE

€ 1 999 860,00

Coordonné par

UNIVERSITY OF YORK
United Kingdom

Periodic Reporting for period 5 - FLEXSEM (Graded constraints in semantic cognition: How do we retrieve knowledge in a flexible way?)

Période du rapport: 2024-04-01 au 2024-09-30

Résumé du contexte et des objectifs généraux du projet

For any concept (e.g. DOG), we have knowledge about diverse features – for example, a dog is furry, can chase rabbits, and is “man’s best friend”. This project examines the brain processes that allow us to flexibly retrieve relevant conceptual knowledge that suits our current goals and context. We can promote coherence between weakly-related aspects of knowledge as required (for example, to come up with creative solutions), and also achieve the timely release from patterns of retrieval when the situation changes. We can also tailor our ongoing retrieval to focus on specific features or associations that are relevant to our current goals - for example, if we are looking for an object we can use as a fan, we need to focus on items that are flat and stiff, even if these are not the most obvious features we know about the concepts. This ability to be flexible with our semantic knowledge is likely to play a central role in our mental lives – yet the underlying processes are poorly understood because past research has largely focused on how the conceptual store captures what is generally true across experiences (i.e. semantic representation). This project aimed to develop a better understanding of the mechanisms that shape and support our thoughts and behaviour through flexible semantic cognition. The project also has relevance for understanding the difficulties of patients with left hemisphere stroke, who often have semantic control impairments which mean they are relatively inflexible in the way they process conceptual knowledge.

We characterised the brain processes underpinning flexible semantic retrieval, in terms of patterns of activation and connectivity across the brain's surface, over time, and in terms of causal mechanisms. Control over semantic retrieval involves the recruitment of additional brain regions and networks which place constraints on patterns of retrieval in the semantic store. We asked if these regions and networks are organised in an orderly fashion along the cortical surface, testing the hypothesis that network transitions in semantic cognition follow a systematic pattern as you move from the default mode network that supports coherent, heteromodal patterns of retrieval, through control regions to attentional networks that can focus processing on specific non-dominant but currently relevant features. Our “graded constraints” hypothesis predicts that the location of networks is non-arbitrary, with brain regions further away from the heteromodal semantic store supporting retrieval when there is a greater mismatch between ongoing retrieval and the pattern required by the context. We tested this idea with complementary neuroscientific methods that characterise functional recruitment with high resolution in space (magnetic resonance imaging) and time (magnetoencephalography). We investigated causality (neuropsychology) and the broader implications of our account for cognition (using an individual differences approach).

Travail effectué depuis le début du projet jusqu’à la fin de la période considérée dans le rapport et principaux résultats atteints jusqu’à présent

Our results support the view that the brain networks that underpin flexible semantic cognition are organised systematically on the cortical surface, supporting our 'graded constraints' account. We obtained evidence to show that semantic control is supported by a large-scale network, juxtaposed between the default mode network and the multiple-demand network that supports domain-general aspects of control. We also obtained convergent evidence across methods to indicate that semantic control is distinct from domain-general executive control.

1: Neuropsychology. We tested participants with semantic aphasia, who have deficient semantic control following left hemisphere stroke. These patients can understand coherent combinations of concepts relatively well but struggle whenever the task requires more control; for example, because it involves semantic ambiguity, weaker associations, or a pattern of retrieval that is at odds from other recently-retrieved features. We examined changes in brain connectivity that are associated with problems with semantic and non-semantic control following stroke. Although these control impairments are associated with lesions to adjacent parts of cortex, they reflect different patterns of structural disconnection. Poorer semantic cognition occurs when left hemsipehere semantic control regions are disconnected, suggesting a distributed yet separate network for semantic control. Poorer non-semantic control was associated with disconnection between the two hemispheres, consistent with the hypothesis that semantic control is dependent on a left-lateralised network, while non-semantic control is more bilateral.

2. Functional MRI. In healthy volunteers, we showed that the semantic control network is partially distinct from the multiple-demand network for domain-general control, and is located between the default mode network and domain-general control regions on the cortical surface in the left hemisphere. We also found that semantic control can operate in top-down and bottom-up modes: when we are trying to identify a creative link between weakly related concepts, or when we have a goal for retrieval in mind, we activate the same semantic control network. However, only top-down semntic control can be acheived by gating input processes, such that semantic activation from words and pictures is more efficently tailored to the task. In contrast, when multiple pieces of information push us towards a consistent pattern of conceptual retrieval, the need for control is reduced and activation is focussed in the default mode network.

3. Time-sensitive magnetoencephalography (MEG). We used MEG to show that the oscillatory response to a pair of words differs across the semantic network, depending on the relationship between the items. When two words have highly coherent meanings, there is a stronger response in the anterior parts of the temporal lobes, close to default mode network areas, relatively late (suggesting coherence may build over time). When the words are only weakly associated, there is a stronger response in posterior temporal areas associated with semantic control, at an earlier time point.

4. Individual differences in semantic cognition. This part of the project explored the functional consequences of the neural architecture we have delineated. We found individual differences within the semantic control network are associated with variability in the efficiency of controlled semantic retrieval and with differences in creativity.

Progrès au-delà de l’état des connaissances et impact potentiel prévu (y compris l’impact socio-économique et les conséquences sociétales plus larges du projet jusqu’à présent)

This work advances our understanding of how flexible semantic retrieval emerges from interactions between control processes and long-term conceptual representations. Our use of cortical gradients to understand the functional organisation of cognition goes beyond the state-of-the-art by showing there are systematic functional transitions between the networks that support semantic cognition in different parts of the brain. We examined a multidimensional state space that takes cortical gradients as dimensions, and showed this neural state space can separate different aspects of semantic cognition. This captures the way that brain networks contribute to different aspects of cognition at different times (for example, default mode regions can support both semantic tasks and off task thought, depending on their relationship with control systems).

fig-1.png