A Systematic Analysis of Linguistic Features in AI-Generated Text Detection Across Domains and Models
Summary
A large-scale empirical study analyzes 284 linguistic features across 27 LLMs and 10 text domains to assess which features reliably detect AI-generated text. The study finds that lexical richness measures are the most robust cross-domain and cross-model signals, while many other proposed indicators are strongly context-dependent.
View Cached Full Text
Cached at: 06/05/26, 02:13 AM
# A Systematic Analysis of Linguistic Features in AI-Generated Text Detection Across Domains and Models Source: [https://arxiv.org/abs/2606.04177](https://arxiv.org/abs/2606.04177) [View PDF](https://arxiv.org/pdf/2606.04177) > Abstract:Interpretable linguistic features offer a promising approach for explaining why a given text appears machine\-generated, particularly for non\-expert users\. However, existing findings on which features reliably indicate LLM\-generated text remain fragmented across feature sets, models, and text domains\. To address this gap, we conduct a large\-scale empirical study assessing the robustness of linguistic signals for characterizing AI\-generated text\. Our analysis covers 284 interpretable linguistic features across outputs from 27 LLMs and ten text domains under cross\-model and cross\-domain generalization settings\. We show that classifiers based solely on linguistic features can reliably distinguish AI\-generated from human\-written text\. However, many previously proposed indicators prove strongly context\-dependent, with the exception of measures of lexical richness, which remain robust signals across model families and text domains\. These results demonstrate which linguistic signals generalize across contexts and provide a foundation for more reliable, interpretable analyses of AI\-generated language\. ## Submission history From: Esra Dönmez \[[view email](https://arxiv.org/show-email/ce7b820c/2606.04177)\] **\[v1\]**Tue, 2 Jun 2026 19:46:22 UTC \(8,530 KB\)
Similar Articles
Findings of the Counter Turing Test: AI-Generated Text Detection
This paper presents findings from the Counter Turing Test shared task on AI-generated text detection, with top systems achieving perfect binary classification but significantly lower performance in model attribution, highlighting the difficulty of distinguishing outputs from different large language models.
DetectRL-X: Towards Reliable Multilingual and Real-World LLM-Generated Text Detection
DetectRL-X is a comprehensive multilingual benchmark for evaluating LLM-generated text detectors across 8 languages and 6 domains, including stress testing with AI-assisted writing operations and perturbations. It reveals strengths and limitations of current detectors in multilingual scenarios.
Cross-Prompt Generalization in Detecting AI-Generated Fake News Using Interpretable Linguistic Features
Researchers from Kennesaw State University investigate cross-prompt generalization in detecting AI-generated fake news using interpretable linguistic features (lexical diversity, readability, emotion). A random forest classifier trained on one prompting strategy and tested on another achieves AUC values of 0.988–1.000, suggesting these features capture stable, generalizable properties of AI-generated text.
How Human-Like Are Large Language Models? A Register-Aware Linguistic Evaluation Framework
This paper introduces a register-aware linguistic evaluation framework to assess how human-like large language models (LLMs) are by comparing the distribution of 67 lexico-grammatical features between human and LLM-generated texts using Maximum Mean Discrepancy. Experiments across seven instruction-tuned open-source models and five registers show that no model perfectly matches human baselines, and closeness to human language varies by register rather than model size.
Multimodal AI Detection for Math-Heavy Scientific Text
该论文研究了在数学密集的科学文本中检测AI生成内容的难题，发现现有检测器（如Pangram 4）在数学内容增多时性能显著下降，并提出基于图像的检测器SnapJudge，在0.1%误报率下对前沿模型生成的证明和散文达到92%和93%的检测率。