Skip to main content

Cortical Representation of Abstract Semantic Knowledge

Periodic Reporting for period 4 - CRASK (Cortical Representation of Abstract Semantic Knowledge)

Reporting period: 2019-11-01 to 2021-07-31

Conceptual representation in the brain has been extensively studied in terms of singular concepts, like an apple (a red/green, round, edible, fruit). The challenge of CRASK is to move beyond this to how we link these single concepts to form our complex encyclopaedic knowledge (the rebellious Swiss, William Tell, once shot an apple off his son's head with a crossbow). CRASK uses magnetoencephalography (MEG) and functional magnetic resonance imaging (fMRI) to meet this challenge. First creating a model of how brain regions work together during the processing of simple concepts, then using this to model how the brain creates the combinatorial world-knowledge that is so important for our daily lives.

World-knowledge plays a big role in our professional, scholastic and interpersonal success. CRASK is motivated by the belief that understanding how this knowledge is instantiated in the brain will allow us to improve our understanding of what makes some individuals better at knowing things than others, why our ability to access our stored knowledge is sometimes transiently blocked and how our ability to access knowledge changes over the lifespan.
In the first part of this project, we have exploited the brain’s propensity towards interactions with conspecifics to characterise how complex knowledge is stored in the brain. In particular, when we are presented with a familiar person, the regions of the brain's 'semantic system' are spontaneously recruited. Taking advantage of this, we used fMRI in a novel way, overloading the system through the rapid presentation of stimuli (famous faces) to determine how long different brain regions are activated during a cognitive process (regional 'time constants'). Knowing how long a brain region takes to perform its cognitive function gives us insight into the depth and complexity of the type of knowledge stored in this region. The results are surprising and interesting. While regions involved in perceiving the faces of other people show a defined hierarchy - with progressively longer time constants in brain regions associated with more sophisticated stages of face perception - regions involved in knowing about other people express a common temporal tuning profile (peaking around .8 seconds when looking at a famous person). This suggests that these distributed brain regions are commonly active and coordinate with each other to endow us the rich and diverse information we have about other humans. In a separate study, we looked at how the coordination between these same brain regions changes when we access a famous person's name, facts about them, social information, personal memories, or physical characteristics. This allowed us to determine that the brain processes and stores factual information and personal memories in a similar way as well as understanding the role that each part of the brain network contributes to the overall picture.
Moving to object concepts in general, we considered whether the mode by which we form knowledge about a category shapes the representation of facts. Specifically, we asked our Italian subjects to retrieve the part of Italy traditional Italian food dishes come from. We observed that a specific part of the brain, the insula, was active when encyclopaedic information was retrieved about food (but not about people or cities). This is interesting as the insula is involved with taste information, but our research seems to suggest that this brain region is co-opted by abstract information that has little to do with taste, as long as it relates to the category 'food'. At the same time, we found that when non-stereotypical geographic information has to be retrieved about food dishes or people (‘where, in Italy, did this item originate’), the brain flexibly recruits cortical regions specialized for other categories of objects. Here, regions classically associated with the processing of places. This is important as it reveals the influence of semantic content and the mechanism by which flexible information can be retrieved about a given object, one of the necessary components of higher-level encyclopaedic knowledge.
In two related studies we asked how elements of the brain’s semantic system contribute to different stage in semantic processing. In one study, we compared automatic access to meaning when reading words (where the task required judgments of the superficial, phonetic, aspect of words, rather than their meaning), to effortful semantic access (where participants made judgments about the semantic typicality of the word within it semantic class). We observed common representations of different semantic classes in the voxel level pattern of activation in the ventrotemporal cortex, posterior middle temporal gyri and lateral inferior frontal gyrus in the brain’s left hemisphere. In a second study, we used cross imaging-modality informational mapping to align fMRI and MEG results to identify the early evocation of semantic representation in these same regions, followed by later semantic representation in ‘default mode’ components of the brain’s semantic system that are implicated in a broad range of internalized cognitive processes.
Having investigated the role of single concepts within the brain, CRASK moved onto the integration of singular concepts into combined, complex semantic knowledge. Employing single sentences drawing on subject/object nouns drawn from five different domains of knowledge (people, places, food, animals & objects), we determined that object domain influences the cortical locus of semantic representation to a greater extent than is generally seen with single concepts, that this persists when sentences draw on nouns from different knowledge-domains and that the linkage across object domains is accompanied by increased activation of the precuneus, a component of the semantic system associated with internalized cognition.
In a series of related works, we built on these findings to investigate how real-world factors relating to complex factual knowledge. Using trivia type questions, again loading on different domains of knowledge, we asked what brain states are associated with temporarily blocked and successful access to the things we know, the successful versus unsuccessful encoding of new knowledge, what differentiates individuals who are better or worse at storing and accessing complex knowledge, and how brain states change as we age from younger to older adults.
CRASK has advanced the state of the art in several directions. The first stage provided new insight into the division of labour within the system for perceiving and knowing about other people, and novel procedures probed temporal processing times to map the hierarchical organisation of this cortical network, while network level multivariate analysis of subtle changes in activation across regions when different forms of knowledge are accessed (e.g. Tell’s rebelliousness, name and nationality) quantified the relationship between access to different forms of knowledge. CRASK showed how non-stereotypical information about an object class can be accomplished in the brain through the co-opting of regions involved in the representation of other object classes and how the automatic access to conceptual knowledge we experience when we see or read the name of an object is differentiated in the brain from when we purposefully access additional information about that concept. CRASK showed how different singular concepts are selectively represented in the brain, and how they may be linked together across these regions through processes occurring in centralised nodes, as well as identifying the neural states that are associated with temporary failures in semantic access, better aptitudes for storing and accessing complex knowledge, and is investigating the neural systems associated with the successful learning of new information as well as changes as we age from younger to older adults.