Periodic Reporting for period 1 - MAGIMAB (Maps And Graphs In Mind And Brain)
Período documentado: 2023-02-01 hasta 2025-01-31
In the Critique of Pure Reason, Immanuel Kant argued that space and time are our a priori senses to perceive the universe. Since then, the quest for how our mind reflects the world and structures our knowledge has never ended, becoming the central theme in psychology, neuroscience, and artificial intelligence. Two major forms of mental representations have gradually captured attention. One is the cognitive map, which Edward Tolman initially proposed to represent the spatial layout of an environment at the dawn of the cognitive revolution. The other is the cognitive graph, which is often utilized to represent the transitions among temporal events.
The discrepancy between the two representations stems from their fundamental differences in representational formats. A cognitive map is typically considered Euclidean, consisting of two or three continuous dimensions. It defines identities as coordinates. These identities could form clusters where the within-cluster Euclidean distance is shorter than the between-cluster one. Algorithms like k-means can help identify the clusters. A cognitive graph, instead, consists of discrete nodes with undirected or directed edges connecting them. It defines identities among relations. These identities could form modules where the within-module connection is stronger than the between-module one. Algorithms based on graph spectrums can help detect the modules.
Through their unique formats, cognitive maps and cognitive graphs are considered to fulfill different roles in physical cognition. While the map-like representation portrays the spatial layout (i.e. the locations of places), the graph-like representation depicts the topological structure (i.e. the routes among places). A directed graph can also represent transitions among states of affairs (i.e. the narrative plot). These mental representations allow us to travel in space and time without physically being there, which offers us enormous survival advantages: we can figure out the optimal plan without exhausting all the possibilities.
Current cutting-edge research and theories have suggested that cognitive maps not only relate to physical space but also can be a general coding mechanism capable of organizing knowledge in a semantic space. However, the unique roles of map-like versus graph-like representations in semantic cognition are still not well articulated. This project tested the hypothesis that the cognitive map and the cognitive graph lay the foundation for the two major approaches to structuring knowledge. One approach assumes concepts are componential, consisting of shared semantic features. Similarity among semantic features leads to taxonomic relations and categories (e.g. “lion” and “tiger” are cats). The other approach assumes concepts are holistic. Co-occurrence among concepts (either in the world or in language) leads to thematic relations and categories (e.g. “lion” and “zebra” co-occur in the savanna). The cognitive map can support feature-based conceptual representations through the correspondence between spatial dimensions and semantic features, whereas the cognitive graph can underly the holistic semantic representations by connecting the concurring concepts.
If this hypothesis is true, the cognitive map and the cognitive graph may further relate to social cognition. On the one hand, evidence suggests that people have different preferences in navigational strategies. About half of the population are spatial navigators who rely more on the cognitive map (e.g. the spatial layout of the landmarks). The other half of the population are narrative navigators who rely more on the cognitive graph (e.g. the topological structure of routes). On the other hand, evidence also indicates individual differences in thinking styles and their correspondence to cultural norms. Individualistic cultures emphasize taxonomies by attending to entities and attributes, whereas collectivistic cultures emphasize themes by attending to contexts and relations. Would a spatial navigator tend to be a taxonomic thinker and an individualist, and a narrative navigator tend to be a thematic thinker and a collectivist? The piece of evidence that would bridge cognitions across three domains is still missing.
Recent decades have seen significant breakthroughs in neuroscience concerning neural coding in both physical and conceptual domains. However, researchers fail to reach a consensus about the neural correlates of the cognitive map and the cognitive graph. The critical issue of these studies is that they ground their hypotheses on one of the representations without considering the other, which leads to a potential confounding between the representation of interest and the alternative representation. Resolving this issue requires studies that can orthogonalize the map and the graph content and pay equal attention to both representations. Such investigations, however, are entirely missing in the literature.
This project aims to disentangle cognitive maps and cognitive graphs in our mind and brain and to build a framework to understand how these two central mental representations support our cognition across physical, conceptual, and cultural domains. It will have the following three specific objectives:
• Objective 1: Disentangling neural correlates of cognitive maps and cognitive graphs in the physical domain.
• Objective 2: Disentangling neural correlates of cognitive maps and cognitive graphs in the conceptual domain (including the semantic and the social domains).
• Objective 3: Investigating the individual preference for cognitive maps and cognitive graphs, bridging cognition across physical, conceptual, and cultural domains.
The discrepancy between the two representations stems from their fundamental differences in representational formats. A cognitive map is typically considered Euclidean, consisting of two or three continuous dimensions. It defines identities as coordinates. These identities could form clusters where the within-cluster Euclidean distance is shorter than the between-cluster one. Algorithms like k-means can help identify the clusters. A cognitive graph, instead, consists of discrete nodes with undirected or directed edges connecting them. It defines identities among relations. These identities could form modules where the within-module connection is stronger than the between-module one. Algorithms based on graph spectrums can help detect the modules.
Through their unique formats, cognitive maps and cognitive graphs are considered to fulfill different roles in physical cognition. While the map-like representation portrays the spatial layout (i.e. the locations of places), the graph-like representation depicts the topological structure (i.e. the routes among places). A directed graph can also represent transitions among states of affairs (i.e. the narrative plot). These mental representations allow us to travel in space and time without physically being there, which offers us enormous survival advantages: we can figure out the optimal plan without exhausting all the possibilities.
Current cutting-edge research and theories have suggested that cognitive maps not only relate to physical space but also can be a general coding mechanism capable of organizing knowledge in a semantic space. However, the unique roles of map-like versus graph-like representations in semantic cognition are still not well articulated. This project tested the hypothesis that the cognitive map and the cognitive graph lay the foundation for the two major approaches to structuring knowledge. One approach assumes concepts are componential, consisting of shared semantic features. Similarity among semantic features leads to taxonomic relations and categories (e.g. “lion” and “tiger” are cats). The other approach assumes concepts are holistic. Co-occurrence among concepts (either in the world or in language) leads to thematic relations and categories (e.g. “lion” and “zebra” co-occur in the savanna). The cognitive map can support feature-based conceptual representations through the correspondence between spatial dimensions and semantic features, whereas the cognitive graph can underly the holistic semantic representations by connecting the concurring concepts.
If this hypothesis is true, the cognitive map and the cognitive graph may further relate to social cognition. On the one hand, evidence suggests that people have different preferences in navigational strategies. About half of the population are spatial navigators who rely more on the cognitive map (e.g. the spatial layout of the landmarks). The other half of the population are narrative navigators who rely more on the cognitive graph (e.g. the topological structure of routes). On the other hand, evidence also indicates individual differences in thinking styles and their correspondence to cultural norms. Individualistic cultures emphasize taxonomies by attending to entities and attributes, whereas collectivistic cultures emphasize themes by attending to contexts and relations. Would a spatial navigator tend to be a taxonomic thinker and an individualist, and a narrative navigator tend to be a thematic thinker and a collectivist? The piece of evidence that would bridge cognitions across three domains is still missing.
Recent decades have seen significant breakthroughs in neuroscience concerning neural coding in both physical and conceptual domains. However, researchers fail to reach a consensus about the neural correlates of the cognitive map and the cognitive graph. The critical issue of these studies is that they ground their hypotheses on one of the representations without considering the other, which leads to a potential confounding between the representation of interest and the alternative representation. Resolving this issue requires studies that can orthogonalize the map and the graph content and pay equal attention to both representations. Such investigations, however, are entirely missing in the literature.
This project aims to disentangle cognitive maps and cognitive graphs in our mind and brain and to build a framework to understand how these two central mental representations support our cognition across physical, conceptual, and cultural domains. It will have the following three specific objectives:
• Objective 1: Disentangling neural correlates of cognitive maps and cognitive graphs in the physical domain.
• Objective 2: Disentangling neural correlates of cognitive maps and cognitive graphs in the conceptual domain (including the semantic and the social domains).
• Objective 3: Investigating the individual preference for cognitive maps and cognitive graphs, bridging cognition across physical, conceptual, and cultural domains.
To meet Objective 1, I have conducted a functional magnetic resonance imaging (fMRI) experiment. In this experiment, I designed a virtual environment where the map content (the pairwise Euclidean Distance) and the graph content (e.g. the pairwise shortest path distance) of the monuments are orthogonal to each other. Sixty participants have been recruited. They remembered the positions of the monuments before scanning and recalled their positions during the scanning. I have conducted the univariate contrast analysis and the parametric modulation analysis to disentangle neural correlates of cognitive maps and cognitive graphs in the physical domain.
To meet Objective 2, I am conducting the second fMRI experiment. I developed a visual novel paradigm to simulate the real-life semantic and social learning process where the map content (i.e. taxonomic relations or personality similarity) and the graph content (i.e. thematic relations or social relations) are orthogonal to each other in the conceptual domain. Since we aim to recruit two groups of participants (one for semantic learning and the other for social learning), the data collection is still ongoing. After data collection, I will measure the map-like and the graph-like neural representations and test whether they are shared, separated, or modulated by the task context. I will also compare the neural correlates of cognitive maps and cognitive graphs for conceptual cognition to those for physical cognition found in the first experiment to test whether the exact neural implementations lay the foundation for our cognition across domains.
To meet Objective 3, I have developed questionnaires and behavioral tests to investigate participants’ tendencies towards being a spatial navigator or a narrative navigator, a taxonomic thinker or a thematic thinker, an individualist or a collectivist. All the data collection procedures were onsite to measure the reaction time better. I have collected data from 158 participants. Sixty participants were from the first fMRI experiment, the rest 98 were independent participants, and the participants in the second fMRI experiment will also participate in the behavioral experiment. The data analysis is still ongoing. I will test whether cognitive maps and graphs support our high-level cognition by examining whether a spatial navigator is more likely to be a taxonomic thinker and an individualist. I will also test whether the brain activity profile in the two above fMRI experiments can predict the participants’ behavior.
To meet Objective 2, I am conducting the second fMRI experiment. I developed a visual novel paradigm to simulate the real-life semantic and social learning process where the map content (i.e. taxonomic relations or personality similarity) and the graph content (i.e. thematic relations or social relations) are orthogonal to each other in the conceptual domain. Since we aim to recruit two groups of participants (one for semantic learning and the other for social learning), the data collection is still ongoing. After data collection, I will measure the map-like and the graph-like neural representations and test whether they are shared, separated, or modulated by the task context. I will also compare the neural correlates of cognitive maps and cognitive graphs for conceptual cognition to those for physical cognition found in the first experiment to test whether the exact neural implementations lay the foundation for our cognition across domains.
To meet Objective 3, I have developed questionnaires and behavioral tests to investigate participants’ tendencies towards being a spatial navigator or a narrative navigator, a taxonomic thinker or a thematic thinker, an individualist or a collectivist. All the data collection procedures were onsite to measure the reaction time better. I have collected data from 158 participants. Sixty participants were from the first fMRI experiment, the rest 98 were independent participants, and the participants in the second fMRI experiment will also participate in the behavioral experiment. The data analysis is still ongoing. I will test whether cognitive maps and graphs support our high-level cognition by examining whether a spatial navigator is more likely to be a taxonomic thinker and an individualist. I will also test whether the brain activity profile in the two above fMRI experiments can predict the participants’ behavior.
Currently, the project has generated the following results:
(1) A virtual environment for testing the hypothesis of cognitive maps and graphs
(2) A learning paradigm using visual novels to simulate real-life conceptual learning and social learning experiences
(3) Brain maps showing the distinct and shared brain systems involved in representations of cognitive maps and cognitive graphs
(1) A virtual environment for testing the hypothesis of cognitive maps and graphs
(2) A learning paradigm using visual novels to simulate real-life conceptual learning and social learning experiences
(3) Brain maps showing the distinct and shared brain systems involved in representations of cognitive maps and cognitive graphs