A unitary representational space for abstract and concrete concepts derived from human similarity judgments
Year: 2026
Type: preprint
Abstract: Human semantic memory is a vast store of knowledge about people, places, and things (i.e., concrete concepts), and about intangible concepts like norms, ideas, and relations (i.e., abstract concepts). A central goal of cognitive science is to understand how concrete and abstract concepts are represented in semantic memory. We used an approximate Bayesian method called Variational Interpretable Concept Embeddings (VICE) to embed the meaning of 751 concrete and abstract nouns in a vector space using over 1.4 million human similarity judgments. Crucially, this method is theory-agnostic, hence the dimensions in the model were not constrained to a specific number or to have predetermined meanings. The result is a 36-dimensional semantic model, in which concrete and abstract concepts are intermingled both within dimensions and across the representational space. The meanings captured by individual dimensions are interpretable and the resultant representational space is organized along two main axes of ‘concreteness’ and ‘agency’. Furthermore, the model can incorporate new data to grow the set of words it is trained on, thus providing a powerful tool for testing theories about the mental organization of human knowledge.
Source: Research Square
Language: English
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7165665483