Scholarlink

This paper studies the detection of Machine-Generated Text (MGT) in Arabic. In this context, we aim to study Arabic MGT detection by constructing (Multi-domain Automatically Generated Texts in Arabic), a comprehensive dataset of diverse human and machine-generated Arabic texts from multiple domains: news, reviews and wiki-how articles. Our approach ensures comparability between human and generated texts, emphasizing their contextual alignment. We additionally train and evaluate various Arabic MGT detectors encompassing both traditional machine learning techniques and state-of-the-art deep models. Our findings further underscore the suitability of Transformers for MGT detection, outperforming other models by considerable margins of over 12 points of Macro-F \(_1\) .

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

MAGENTA: Generating and Detecting Arabic Machine-Generated Text in Multiple Domains

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Saad Yaquine,
  • Amine Hmimou,
  • Paolo Rosso

摘要

This paper studies the detection of Machine-Generated Text (MGT) in Arabic. In this context, we aim to study Arabic MGT detection by constructing (Multi-domain Automatically Generated Texts in Arabic), a comprehensive dataset of diverse human and machine-generated Arabic texts from multiple domains: news, reviews and wiki-how articles. Our approach ensures comparability between human and generated texts, emphasizing their contextual alignment. We additionally train and evaluate various Arabic MGT detectors encompassing both traditional machine learning techniques and state-of-the-art deep models. Our findings further underscore the suitability of Transformers for MGT detection, outperforming other models by considerable margins of over 12 points of Macro-F \(_1\) .