Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster C in Poster Session C: Wednesday, August 5, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

A Compositional Model of Semantic Fluency

Surabhi S Nath¹, Alireza Modirshanechi², Peter Dayan¹; ¹Max Planck Institute for Biological Cybernetics, ²Helmholtz Zentrum München

Presenter: Surabhi S Nath

The ability to recall semantically connected concepts---be it animals, fruits, or capital cities---is a remarkable capacity of the human mind. Such semantic fluency is thought to rely on traversing a mental space in which concepts are represented by their meanings. However, the structure, properties, and navigability of this representational space remain enigmatic and are highly debated. Existing approaches rely either on complex, uninterpretable distributional word-embeddings or on rigid, hand-crafted category norms. Here, we exploit the strengths of both, introducing Conceptome: a compositional, interpretable, feature-based representation of semantic concepts, constructed using large language models. Using the compositional structure of Conceptome, we develop an autoregressive model, Conceptome-search, that characterises how humans traverse their mental space to retrieve concepts. Applied to an animal fluency task, Conceptome-search predicts human choices better than state-of-the-art models, offering new insights into the mechanisms underlying semantic fluency.

Topic Area: Memory, Learning & Knowledge Structures