Saved in:
Bibliographic Details
Main Authors: Chen, Michelle Chao, Miller, Moritz, Schölkopf, Bernhard, Guo, Siyuan
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.17869
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910000726671360
author Chen, Michelle Chao
Miller, Moritz
Schölkopf, Bernhard
Guo, Siyuan
author_facet Chen, Michelle Chao
Miller, Moritz
Schölkopf, Bernhard
Guo, Siyuan
contents Learning structural information from observational data is central to producing new knowledge outside the training corpus. This holds for mechanistic understanding in scientific discovery as well as flexible test-time compositional generation. We thus study how language models learn abstract structures and utilize the learnt structural information at test-time. To ensure a controlled setup, we design a natural language dataset based on linguistic structural transformations. We empirically show that the emergence of learning structural information correlates with complex reasoning tasks, and that the ability to perform test-time compositional generation remains limited.
format Preprint
id arxiv_https___arxiv_org_abs_2601_17869
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle On the Emergence and Test-Time Use of Structural Information in Large Language Models
Chen, Michelle Chao
Miller, Moritz
Schölkopf, Bernhard
Guo, Siyuan
Computation and Language
Machine Learning
Learning structural information from observational data is central to producing new knowledge outside the training corpus. This holds for mechanistic understanding in scientific discovery as well as flexible test-time compositional generation. We thus study how language models learn abstract structures and utilize the learnt structural information at test-time. To ensure a controlled setup, we design a natural language dataset based on linguistic structural transformations. We empirically show that the emergence of learning structural information correlates with complex reasoning tasks, and that the ability to perform test-time compositional generation remains limited.
title On the Emergence and Test-Time Use of Structural Information in Large Language Models
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2601.17869