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

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: McGovern, Hope, Craig, Caroline, Lippincott, Thomas, Sirin, Hale
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914446242217984
author McGovern, Hope
Craig, Caroline
Lippincott, Thomas
Sirin, Hale
author_facet McGovern, Hope
Craig, Caroline
Lippincott, Thomas
Sirin, Hale
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2604_03877
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle When Models Know More Than They Say: Probing Analogical Reasoning in LLMs
McGovern, Hope
Craig, Caroline
Lippincott, Thomas
Sirin, Hale
Computation and Language
Artificial Intelligence
Machine Learning
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.
title When Models Know More Than They Say: Probing Analogical Reasoning in LLMs
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2604.03877