From Associations to Activations: Comparing Behavioral and Hidden-State Semantic Geometry in LLMs

2026-01-31

AI Summary

This study investigates whether the behavioral outputs of large language models (LLMs) reflect the semantic geometry of their internal hidden states. Using psycholinguistic experiments—specifically forced-choice and free-association tasks—the authors collect behavioral data from eight instruction-tuned Transformer models across 5,000 words, constructing behavioral similarity matrices. These are then compared with hidden-state representations at each layer via representational similarity analysis. The work presents the first systematic comparison of the two behavioral paradigms in recovering LLMs’ internal semantic geometry, revealing that forced-choice significantly outperforms free association. Notably, behavioral similarities derived from forced-choice effectively predict hidden-state similarities for unseen words, surpassing lexical baselines such as FastText and BERT, as well as cross-model consensus, thereby demonstrating a strong alignment between behavioral data and internal representations.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We investigate the extent to which an LLM's hidden-state geometry can be recovered from its behavior in psycholinguistic experiments. Across eight instruction-tuned transformer models, we run two experimental paradigms -- similarity-based forced choice and free association -- over a shared 5,000-word vocabulary, collecting 17.5M+ trials to build behavior-based similarity matrices. Using representational similarity analysis, we compare behavioral geometries to layerwise hidden-state similarity and benchmark against FastText, BERT, and cross-model consensus. We find that forced-choice behavior aligns substantially more with hidden-state geometry than free association. In a held-out-words regression, behavioral similarity (especially forced choice) predicts unseen hidden-state similarities beyond lexical baselines and cross-model consensus, indicating that behavior-only measurements retain recoverable information about internal semantic geometry. Finally, we discuss implications for the ability of behavioral tasks to uncover hidden cognitive states.

Problem

• semantic geometry
• large language models
• behavioral experiments
• hidden states
• representational similarity

Innovation

• representational similarity analysis
• behavioral semantics
• hidden-state geometry
• forced-choice task
• large language models