The Grounding Gap: How LLMs Anchor the Meaning of Abstract Concepts Differently from Humans
On the map: Mechanistic interpretability → Knowledge and representations inside LLMs
Abstract (arXiv)
Abstract concepts - justice, theory, availability - have no single perceivable referent; in the human brain, their meaning emerges from a web of experiences, affect, and social context. Do large language models (LLMs) ground abstract concepts in a similar way? We study this by replicating property-generation experiments from cognitive science on 21 frontier and open-weight LLMs. Across models and experiments, we find a consistent pattern: when compared to humans, models rely too heavily on word associations, and underproduce properties tied to emotion and internal states. This yields a large and consistent grounding gap: no model exceeds a Pearson correlation r=0.37 with human responses, compared to a human-to-human ceiling above r=0.9. To better interpret this gap, we also replicate a rating experiment on grounding categories and find that here LLMs align more closely with human judgment, and alignment improves as models get larger. We then use sparse autoencoders (SAEs) to inspect whether this information is also reflected in the models' internal features, and we do identify features connected to grounding dimensions such as "sensorimotor" and "social". These findings suggest that current LLMs can recover grounding dimensions when explicitly queried, but do not recruit them in a human-like way when words are generated freely.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Knowledge and representations inside LLMs, one of 13 directions in Mechanistic interpretability. Builds on 17 of its 47 references that are in the map. First posted on arXiv 9 May 2026.
Builds on
17 of 47 references.
References · 17
- 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
- 2022Training Compute-Optimal Large Language Modelscited by 3,019
- 2022Emergent Abilities of Large Language Modelscited by 2,512
- 2023Sparse Autoencoders Find Highly Interpretable Features in Language Modelscited by 1,420
- 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,085
- 2023Steering Language Models With Activation Engineeringcited by 922
- 2022Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Taskcited by 431
- 2024Gemma Scope: Open Sparse Autoencoders Everywhere All At Once on Gemma 2cited by 394
- 2019What BERT is not: Lessons from a new suite of psycholinguistic diagnostics for language modelscited by 320
- 2023Emergent Linear Representations in World Models of Self-Supervised Sequence Modelscited by 319
- 2023Language Models Represent Space and Timecited by 316
- 2021Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Colorcited by 119
5 more
- 2025Embodied AI Agents: Modeling the Worldcited by 61
- 2026Emotion Concepts and their Function in a Large Language Modelcited by 57
- 2026Incompressible Knowledge Probes: Estimating Black-Box LLM Parameter Counts via Factual Capacitycited by 17
- 2023Conceptual structure coheres in human cognition but not in large language modelscited by 14
- 2025AI shares emotion with humans across languages and culturescited by 6
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026Beyond Hooking Onto the World: Referential Profiles and the Numerical Structure of LLM Grounding
- 2024"Understanding AI": Semantic Grounding in Large Language Models
- 2025How Deep is Love in LLMs' Hearts? Exploring Semantic Size in Human-like Cognition
- 2025The Representational Alignment between Humans and Language Models is implicitly driven by a Concreteness Effect
- 2025The Mechanistic Emergence of Symbol Grounding in Language Models
- 2023Grounding Gaps in Language Model Generations
- 2026Do LLMs Have Values? A Quantitative Analysis and Alignment Framework for Values in Large Language Models
- 2025The Homogenizing Effect of Large Language Models on Human Expression and Thought