Discourse Features Enhance Detection of Document-Level Machine-Generated Content

TL;DR

This paper introduces discourse-based features and a new model, DTransformer, to improve detection of machine-generated content at the document level, significantly outperforming existing methods.

Contribution

It presents novel discourse-aware methodologies and datasets, including DTransformer, to enhance detection of machine-generated texts, especially longer and paraphrased documents.

Findings

15.5% absolute improvement on paraLFQA dataset

4% absolute improvement on paraWP dataset

1.5% absolute improvement on M4 dataset

Abstract

The availability of high-quality APIs for Large Language Models (LLMs) has facilitated the widespread creation of Machine-Generated Content (MGC), posing challenges such as academic plagiarism and the spread of misinformation. Existing MGC detectors often focus solely on surface-level information, overlooking implicit and structural features. This makes them susceptible to deception by surface-level sentence patterns, particularly for longer texts and in texts that have been subsequently paraphrased. To overcome these challenges, we introduce novel methodologies and datasets. Besides the publicly available dataset Plagbench, we developed the paraphrased Long-Form Question and Answer (paraLFQA) and paraphrased Writing Prompts (paraWP) datasets using GPT and DIPPER, a discourse paraphrasing tool, by extending artifacts from their original versions. To better capture the structure of longer texts at document level, we propose DTransformer, a model that integrates discourse analysis through PDTB preprocessing to encode structural features. It results in substantial performance gains across both datasets - 15.5% absolute improvement on paraLFQA, 4% absolute improvement on paraWP, and 1.5% absolute improvemene on M4 compared to SOTA approaches. The data and code are available at: https://github.com/myxp-lyp/Discourse-Features-Enhance-Detection-of-Document-Level-Machine-Generated-Content.git.