Probing Syntax in Large Language Models: Successes and Remaining Challenges

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Diego-Simón, Pablo J., Chemla, Emmanuel, King, Jean-Rémi, Lakretz, Yair
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915436635881472
author Diego-Simón, Pablo J.
Chemla, Emmanuel
King, Jean-Rémi
Lakretz, Yair
author_facet Diego-Simón, Pablo J.
Chemla, Emmanuel
King, Jean-Rémi
Lakretz, Yair
contents The syntactic structures of sentences can be readily read-out from the activations of large language models (LLMs). However, the ``structural probes'' that have been developed to reveal this phenomenon are typically evaluated on an indiscriminate set of sentences. Consequently, it remains unclear whether structural and/or statistical factors systematically affect these syntactic representations. To address this issue, we conduct an in-depth analysis of structural probes on three controlled benchmarks. Our results are three-fold. First, structural probes are biased by a superficial property: the closer two words are in a sentence, the more likely structural probes will consider them as syntactically linked. Second, structural probes are challenged by linguistic properties: they poorly represent deep syntactic structures, and get interfered by interacting nouns or ungrammatical verb forms. Third, structural probes do not appear to be affected by the predictability of individual words. Overall, this work sheds light on the current challenges faced by structural probes. Providing a benchmark made of controlled stimuli to better evaluate their performance.
format Preprint
id arxiv_https___arxiv_org_abs_2508_03211
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Probing Syntax in Large Language Models: Successes and Remaining Challenges
Diego-Simón, Pablo J.
Chemla, Emmanuel
King, Jean-Rémi
Lakretz, Yair
Computation and Language
The syntactic structures of sentences can be readily read-out from the activations of large language models (LLMs). However, the ``structural probes'' that have been developed to reveal this phenomenon are typically evaluated on an indiscriminate set of sentences. Consequently, it remains unclear whether structural and/or statistical factors systematically affect these syntactic representations. To address this issue, we conduct an in-depth analysis of structural probes on three controlled benchmarks. Our results are three-fold. First, structural probes are biased by a superficial property: the closer two words are in a sentence, the more likely structural probes will consider them as syntactically linked. Second, structural probes are challenged by linguistic properties: they poorly represent deep syntactic structures, and get interfered by interacting nouns or ungrammatical verb forms. Third, structural probes do not appear to be affected by the predictability of individual words. Overall, this work sheds light on the current challenges faced by structural probes. Providing a benchmark made of controlled stimuli to better evaluate their performance.
title Probing Syntax in Large Language Models: Successes and Remaining Challenges
topic Computation and Language
url https://arxiv.org/abs/2508.03211