ICLR Oral 7: AI-enhanced semantic feature norms for 786 concepts

Semantic feature norms have been foundational in the study of human conceptualknowledge, yet traditional methods face trade-offs between concept/featurecoverage and verifiability of quality due to the labor-intensive nature of normingstudies. Here, we introduce a novel approach that augments a dataset of human-generatedfeature norms with responses from large language models (LLMs) whileverifying the quality of norms against reliable human judgments. We find that ourAI-enhanced feature norm dataset shows much higher feature density and overlapamong concepts while outperforming a comparable human-only norm dataset andword-embedding models in predicting people’s semantic similarity judgments.Taken together, we demonstrate that human conceptual knowledge is richer thancaptured in previous norm datasets and show that, with proper validation, LLMscan serve as powerful tools for cognitive science research.