Saved in:
Bibliographic Details
Main Authors: Li, Yupei, Milling, Manuel, Specia, Lucia, Schuller, Björn W.
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2412.12679
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912844740558848
author Li, Yupei
Milling, Manuel
Specia, Lucia
Schuller, Björn W.
author_facet Li, Yupei
Milling, Manuel
Specia, Lucia
Schuller, Björn W.
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2412_12679
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Discourse Features Enhance Detection of Document-Level Machine-Generated Content
Li, Yupei
Milling, Manuel
Specia, Lucia
Schuller, Björn W.
Computation and Language
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.
title Discourse Features Enhance Detection of Document-Level Machine-Generated Content
topic Computation and Language
url https://arxiv.org/abs/2412.12679