MADOran: A morphologically annotated dataset of Oran

Area of Science:

• Computational Linguistics
• Natural Language Processing
• Arabic Dialectology

Background:

• Limited annotated resources exist for Arabic dialects, hindering NLP development.
• Orani Arabic (ORN) presents unique morphological and syntactic features requiring specialized datasets.
• Existing annotation tools and guidelines need adaptation for dialectal Arabic.

Purpose of the Study:

• Introduce MADOran, a novel morphologically annotated dataset for Orani Arabic.
• Provide a valuable resource for NLP applications and linguistic research on Arabic dialects.
• Facilitate the development of dialect-specific tools and resources for ORN.

Main Methods:

• Collected 30,919 words from diverse written and spoken genres of Orani Arabic.
• Manually annotated each word with a fine-grained tagset (POS, root, pattern, translations).
• Utilized the Dialectal Word Annotation Tool for Arabic (DIWAN), adapting guidelines from The Dynamic Arabella Corpus (Arabella).

Main Results:

• Developed MADOran, a comprehensive dataset with detailed morphological annotations for ORN.
• The dataset covers various topics and conversational contexts, reflecting real-world language use.
• Annotations include part-of-speech, root, pattern, and English/French translations.

Conclusions:

• MADOran is a significant resource for advancing Arabic dialect processing, particularly for ORN.
• The dataset enables training language models and supports comparative linguistic studies with Modern Standard Arabic (MSA).
• MADOran adheres to data stewardship principles, ensuring findability, accessibility, interoperability, and reusability.