How Order-Sensitive Are LLMs? OrderProbe for Deterministic Structural Reconstruction

TL;DR

This paper introduces OrderProbe, a benchmark for assessing how well large language models can reconstruct the original structure of scrambled Chinese, Japanese, and Korean expressions, revealing challenges in structural understanding beyond semantic accuracy.

Contribution

The paper presents OrderProbe, a novel deterministic benchmark and diagnostic framework for evaluating LLMs' ability to reconstruct internal structure from scrambled inputs.

Findings

Structural reconstruction remains difficult for current LLMs, with recovery often below 35%.

Semantic recall and structural planning are dissociated, indicating separate capabilities.

Structural robustness is not an automatic consequence of semantic competence.

Abstract

Large language models (LLMs) excel at semantic understanding, yet their ability to reconstruct internal structure from scrambled inputs remains underexplored. Sentence-level restoration is ill-posed for automated evaluation because multiple valid word orders often exist. We introduce OrderProbe, a deterministic benchmark for structural reconstruction using fixed four-character expressions in Chinese, Japanese, and Korean, which have a unique canonical order and thus support exact-match scoring. We further propose a diagnostic framework that evaluates models beyond recovery accuracy, including semantic fidelity, logical validity, consistency, robustness sensitivity, and information density. Experiments on twelve widely used LLMs show that structural reconstruction remains difficult even for frontier systems: zero-shot recovery frequently falls below 35%. We also observe a consistent dissociation between semantic recall and structural planning, suggesting that structural robustness is not an automatic byproduct of semantic competence.