Steven T. Piantadosi

The paper investigates how semantic information is encoded in human hippocampal activity during speech processing, finding that distributed neural patterns capture semantic relationships similar to language models.

The paper investigates the relationship between logical reasoning and language by examining neural activity and patient performance, finding that logical reasoning does not rely on the language brain network.

The paper identifies simplex-shaped neural manifolds for number representation in the human medial temporal lobe and finds similarities with large language models.

The paper shows that semantic features in word meaning are encoded in the brain through geometric structures similar to those in language models.

The paper identifies a human brain network specialized for abstract formal reasoning using fMRI and distinguishes it from other cognitive networks.

The paper shows that hippocampal neurons encode word position and context in a way similar to self-attention in large language models.

The paper tests whether the language-selective network responds to sentence-level composition despite word order violations in naturalistic sentences.

Monkeys predict future prey positions during pursuit, and dACC neurons encode these predictions.

The paper introduces a concept learning paradigm using term rewriting systems to model human learning of complex conceptual relationships.

Rhesus macaques predict virtual prey trajectories using Newtonian physics principles, ignoring other behavioral patterns despite training.

The paper introduces a Bayesian method to classify neurons into clusters based on response properties, finding that neurons fall into two clusters: task-sensitive and non-selective.

The paper shows that striatum responses reflect a surprise signal from a Bayesian rule-learning model, not reward prediction error, during explicit rule learning.