When Models Know More Than They Say: Probing Analogical Reasoning in LLMs

2026-04-04

AI Summary

Large language models struggle with analogical reasoning when surface cues conflict with underlying structural relationships, revealing limitations in their capacity for abstract generalization. This study systematically evaluates open-source large language models on narrative and rhetorical analogy tasks using both representational probing and standard prompting methods. The findings show that both approaches perform poorly and comparably on narrative analogies, yet probing significantly outperforms prompting on rhetorical analogies. This task-dependent asymmetry between internal representations and overt behavior underscores the unique advantage of probing techniques in uncovering latent knowledge within models. The results offer a novel perspective on the mechanisms underlying analogical reasoning in large language models and highlight the importance of representation-level analysis in assessing their cognitive capabilities.

Technology Category

Cognitive Modeling & Cognitive Systems: AnalogyNatural Language Processing: (Large) Language ModelsMachine Learning: Large Multimodal Models (LMMs)

Application Category

Search and Retrieval-Augmented AI: Large language models for searchUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Analogical reasoning is a core cognitive faculty essential for narrative understanding. While LLMs perform well when surface and structural cues align, they struggle in cases where an analogy is not apparent on the surface but requires latent information, suggesting limitations in abstraction and generalisation. In this paper we compare a model's probed representations with its prompted performance at detecting narrative analogies, revealing an asymmetry: for rhetorical analogies, probing significantly outperforms prompting in open-source models, while for narrative analogies, they achieve a similar (low) performance. This suggests that the relationship between internal representations and prompted behavior is task-dependent and may reflect limitations in how prompting accesses available information.

Problem

• analogical reasoning
• large language models
• narrative understanding
• abstraction
• generalisation

Innovation

• analogical reasoning
• probing
• prompting
• large language models
• narrative understanding