# DependencyAI: Detecting AI Generated Text through Dependency Parsing

Source: [DependencyAI: Detecting AI Generated Text through Dependency Parsing](<https://www.sota2.com/research/paper/6995478c09c33900fcd6d5da>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-19T05:00:04.488Z

Authors: Sara Ahmed, Tracy Hammond

Published: 2026-02-17T11:42:28.000Z

## Abstract

As large language models (LLMs) become increasingly prevalent, reliable methods for detecting AI-generated text are critical for mitigating potential risks. We introduce DependencyAI, a simple and interpretable approach for detecting AI-generated text using only the labels of linguistic dependency relations. Our method achieves competitive performance across monolingual, multi-generator, and multilingual settings. To increase interpretability, we analyze feature importance to reveal syntactic structures that distinguish AI-generated from human-written text. We also observe a systematic overprediction of certain models on unseen domains, suggesting that generator-specific writing styles may affect cross-domain generalization. Overall, our results demonstrate that dependency relations alone provide a robust signal for AI-generated text detection, establishing DependencyAI as a strong linguistically grounded, interpretable, and non-neural network baseline.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [AI-generated text detection on M4GT-Bench Chinese (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-chinese-test.md>): AI-generated text detection; M4GT-Bench Chinese (test); Precision: 93.48.

- [AI-generated text detection on M4GT-Bench English (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-english-test.md>): AI-generated text detection; M4GT-Bench English (test); Precision: 98.02.

- [AI-generated text detection on M4GT-Bench German (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-german-test.md>): AI-generated text detection; M4GT-Bench German (test); Precision: 98.27.

- [AI-generated text detection on M4GT-Bench Russian (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-russian-test.md>): AI-generated text detection; M4GT-Bench Russian (test); Precision: 99.5.

- [AI-generated text detection on M4GT-Bench All (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-all-test.md>): AI-generated text detection; M4GT-Bench All (test); Precision: 96.02.

- [AI-generated text detection on M4GT-Bench PeerRead (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-peerread-test.md>): AI-generated text detection; M4GT-Bench PeerRead (test); Precision: 62.68.

- [AI-generated text detection on M4GT-Bench OUTFOX (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-outfox-test.md>): AI-generated text detection; M4GT-Bench OUTFOX (test); Precision: 54.6.

- [AI-generated text detection on M4GT-Bench Italian (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-italian-test.md>): AI-generated text detection; M4GT-Bench Italian (test); Precision: 96.42.

- [AI-generated text detection on M4GT-Bench WikiHow (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-wikihow-test.md>): AI-generated text detection; M4GT-Bench WikiHow (test); Precision: 61.28.

- [AI-generated text detection on M4GT-Bench Wikipedia (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-wikipedia-test.md>): AI-generated text detection; M4GT-Bench Wikipedia (test); Precision: 44.45.

- [AI-generated text detection on M4GT-Bench Reddit (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-reddit-test.md>): AI-generated text detection; M4GT-Bench Reddit (test); Accuracy: 60.87.

- [AI-generated text detection on M4GT-Bench arXiv (test)](<https://www.sota2.com/research/sota/ai-generated-text-detection-on-m4gt-bench-arxiv-test.md>): AI-generated text detection; M4GT-Bench arXiv (test); Accuracy: 28.53.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2602.15514>)

[PDF](<https://arxiv.org/pdf/2602.15514.pdf>)