Same Meaning, Different Scores: Lexical and Syntactic Sensitivity in LLM Evaluation

TL;DR

This paper investigates how small lexical and syntactic changes in input prompts significantly impact the performance and ranking of large language models, revealing their reliance on surface-level patterns over linguistic understanding.

Contribution

It introduces controlled perturbation pipelines to systematically evaluate LLM robustness, highlighting vulnerabilities and the need for improved evaluation standards.

Findings

Lexical perturbations cause significant performance drops across models.

Syntactic changes have mixed effects, sometimes improving results.

Model robustness does not scale consistently with size.

Abstract

The rapid advancement of Large Language Models (LLMs) has established standardized evaluation benchmarks as the primary instrument for model comparison. Yet, their reliability is increasingly questioned due to sensitivity to shallow variations in input prompts. This paper examines how controlled, truth-conditionally equivalent lexical and syntactic perturbations affect the absolute performance and relative ranking of 23 contemporary LLMs across three benchmarks: MMLU, SQuAD, and AMEGA. We employ two linguistically principled pipelines to generate meaning-preserving variations: one performing synonym substitution for lexical changes, and another using dependency parsing to determine applicable syntactic transformations. Results show that lexical perturbations consistently induce substantial, statistically significant performance degradation across nearly all models and tasks, while syntactic perturbations have more heterogeneous effects, occasionally improving results. Both perturbation types destabilize model leaderboards on complex tasks. Furthermore, model robustness did not consistently scale with model size, revealing strong task dependence. Overall, the findings suggest that LLMs rely more on surface-level lexical patterns than on abstract linguistic competence, underscoring the need for robustness testing as a standard component of LLM evaluation.