Scholarlink

<p>Spelling and grammar errors pose substantial challenges to Arabic text processing due to the language’s complexity, diversity of dialects, and lack of standardized rules. These errors not only degrade readability but also complicate applications in search engines, text analytics, and machine translation. We introduce AraCorrect, an automated Arabic correction system that integrates a transformer-based model, dictionary-based approach, and masked language model in a hybrid pipeline to effectively address diverse error types–spanning spelling, punctuation, segmentation, and grammar. Evaluated on synthetic and real-world datasets, including QALB-2014, AraCorrect achieves significant reductions in Word Error Rate (WER) and demonstrates robust performance, especially in high-error scenarios. Comparative analysis reveals AraCorrect’s effectiveness and reliability over existing tools, making it a robust solution for Arabic text correction across varied applications. AraCorrect leverages a hybrid design to address the contextual rigidity of rule-based systems and the low explainability of deep models.</p>

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

AraCorrect: a hybrid transformer-lexicon approach for enhanced Arabic spelling and grammar correction

  • 期刊论文

  • 发表日期：

    2026年7月17日

查看全文
  • Ensaf Hussein Mohamed,
  • Amr Emam Mohamed,
  • Mariam S. El-Rahmany

摘要

Spelling and grammar errors pose substantial challenges to Arabic text processing due to the language’s complexity, diversity of dialects, and lack of standardized rules. These errors not only degrade readability but also complicate applications in search engines, text analytics, and machine translation. We introduce AraCorrect, an automated Arabic correction system that integrates a transformer-based model, dictionary-based approach, and masked language model in a hybrid pipeline to effectively address diverse error types–spanning spelling, punctuation, segmentation, and grammar. Evaluated on synthetic and real-world datasets, including QALB-2014, AraCorrect achieves significant reductions in Word Error Rate (WER) and demonstrates robust performance, especially in high-error scenarios. Comparative analysis reveals AraCorrect’s effectiveness and reliability over existing tools, making it a robust solution for Arabic text correction across varied applications. AraCorrect leverages a hybrid design to address the contextual rigidity of rule-based systems and the low explainability of deep models.