From cognitive maps to cognitive graphs

Abstract

We investigate the structure of spatial knowledge that spontaneously develops during free exploration of a novel environment. We present evidence that this structure is similar to a labeled graph: a network of topological connections between places, labeled with local metric information. In contrast to route knowledge, we find that the most frequent routes and detours to target locations had not been traveled during learning. Contrary to purely topological knowledge, participants typically traveled the shortest metric distance to a target, rather than topologically equivalent but longer paths. The results are consistent with the proposal that people learn a labeled graph of their environment.

Figure 1. Illustration of three levels of spatial knowledge.

a) Graph knowledge: purely topological graph of a network of place nodes (identifiable places, including junctions) linked by path edges (traversable paths between nodes), expressing the known connectivity of the environment. b) Labeled graph: incorporates local metric information about distances between known places (edge weights) and/or angles between known paths (node labels). Note that the topological structure of the connections between nodes is the same for a) and b). In the labeled graph, metric information may be coarse, contain biases, and is not globally consistent. c) Survey knowledge: configural map-like knowledge of environmental locations. Metric information is quite accurate and consistent throughout the region, embedded in a common coordinate system.

Figure 2. Virtual hedge maze.

a) Overhead view of the maze, which participants never saw. Eight target objects (blue circles) and 4 landmark paintings (red rectangles) were placed in the maze. This figure shows an example of one of the object pairs: a direct trial from the sink (top) to the bookcase (bottom), illustrating the shortest path (solid black line), a topologically equivalent but longer path (dashed orange line), and a path that is both topologically and metrically longer (dotted purple line). Nodes in the paths are indicated by black circles. Thick black arrows indicate that all alternative paths started from the same location next to the sink, and ended by going into the branch hallway of the bookcase. b) View of the VENLab. c) View of landmark inside the maze. d) View of the barrier on a detour trial.

Figure 3. Novel routes taken during test.

An example of a trial starting at the sink (S) and ending at the bookcase (B). a) All paths taken by a representative participant during the 10-minute exploration. b) Those exploration paths that start either from the sink (green, purple) or from the bookcase (orange, blue). c) A novel detour taken during test. The participant never traveled on that path or the reverse path during exploration.

Figure 4. Percentage of trials in which a novel path was taken.

Route knowledge predicts that none of the paths would be novel. Error bars indicate standard error of the mean. *** indicates p<0.001, 1-sample t-test with 0.