Now Reading
Maps of conceptual spaces in the hippocampus

Maps of conceptual spaces in the hippocampus

How are neural representations of conceptual information structured such that people may deduce relationships they have never seen or classify fresh examples? It has been argued that a viable format for encoding physical space during navigation is a representation that resembles a map. Previous research showed distance mapping in a feature space that was significant for concept learning as well as directional coding during navigation across a continuous stimulus feature space. In contrast to a broad feature-based environment, Stephanie Theves et al. present the first evidence in their study for a hippocampus representation of a conceptual space.

We provide the first unambiguous evidence for a hippocampal representation of the actual concept space, by showing that the hippocampal distance signal selectively reflects the mapping of specifically conceptually relevant rather than of all feature dimensions. 

The researchers displayed common items that have previously been linked with particular values on three continuous feature dimensions while fMRI scanning 32 human subjects (21 females). Importantly, prior concept learning was only important across two dimensions. In contrast to distances in a space defined along all feature dimensions, they discovered that hippocampus responses to the objects reflect their relative distances in a space defined along cognitively significant dimensions. According to these results, the hippocampus aids in the acquisition of knowledge by dynamically storing data in a space that is spanned along dimensions that are important for defining ideas.

Experimental design. A, Learning (for details, see B) was framed by OVBs inside the scanner to measure emerging neural representations (for details, see C). A final 3D recall test probed memory across feature dimensions (for details, see D). B, Between OVBs, the following four learning tasks were completed (order: left to right): (1) participants learned six associations between objects and three-dimensional stimuli via encoding and test blocks, and (2) freely recalled them by reconstructing the 3D stimulus associated with a given object until correct; (3) subsequently, they acquired the concept of two stimulus categories via feedback-based categorization along the diagonal in two-dimensional concept space; and (4) navigating concept space required the 2D reconstruction (third dimension randomized) of an object-associated stimulus of a certain category. C, In the OVBs, the six objects plus an additional catch object were presented pseudorandomly with a stimulus duration of 1 s and interstimulus intervals of 3.5, 5, and 6.5 s while participants performed a catch object detection task. D, During post-experimental assessment of memory across feature dimensions, participants had to reconstruct the 3D stimulus associated with a given object and confirm their choice. © 2022, Stephanie Theves et al

Hippocampal distance code for concept space revealed by fMRI adaptation. A, Schematic of two-dimensional object positions and distances between objects in concept space (left) and three-dimensional object positions and distances between objects in feature space (right). B, Average of contrast of parameter estimates (cope) of the 2D versus 3D distance adaptation regressors in all hippocampal voxels. Hippocampal adaptation decreases with the increasing two-dimensional conceptual distance between successively presented objects significantly more than with three-dimensional feature-based distance between objects. C, Control for complexity difference between two- and three-dimensional representation: two-dimensional distances from conceptual space (xy) were compared with two-dimensional distances derived from a combination of the conceptually relevant x-axis (left) and y-axis (right) with the irrelevant z-axis. If the better fit of 2D(xy) distances versus 3D distances (B) reflects a preference of the hippocampus for 2D codes, 2D(xy) should not fit better than 2D(xz) or 2D(yz). Bars reflect the mean. Central marks of the boxes indicate the median, the bottom and top edges of the boxes indicate the 25th and 75th percentiles, whiskers extend to extreme data points not considered outliers; outliers are plotted as red crosses. Asterisk (*) indicates significance at p < 0.05. © 2022, Stephanie Theves et al

Their architecture allowed for the description of object connections in both a three-dimensional feature space defined along all stimulus dimensions and a two-dimensional concept space defined along just conceptually important stimulus dimensions. They discovered that hippocampus representations of things learned previously mirrored their conceptual distances rather than their feature-based distances. The two-dimensional distances in concept space also significantly outperformed alternative two-dimensional distances derived from combinations with the conceptually irrelevant feature dimension in their ability to explain the hippocampal signal, ruling out the possibility that this effect was caused by a hippocampus that prefers two dimensions for coding.

The hippocampal signal thus reflects only a representation of distances in a space spanned by the dimensions that were relevant in relation to one another to define the concept, while the mnemonically relevant third dimension was not integrated in a multidimensional representation.

See Also

This supports the argument that the hippocampus organizes new information in a map-like representation in support of concept learning and it can carve out (and represent) conceptual information from the totality of features, despite encoding specific exemplars in all detail.

The Hippocampus Maps Concept Space, Not Feature Space. Stephanie Theves, Guillén Fernández and Christian F. Doeller.

Published: September 2020
DOI: 10.1523/JNEUROSCI.0494-20.2020

Scroll To Top