Leveled Reading Corpus of Modern Standard Arabic

Introduction

Modern Standard Arabic (MSA) serves as the lingua franca across the Arab world, used in education, media, and formal communication. Despite its widespread use, resources that support progressive reading instruction in MSA remain limited. A Leveled Reading Corpus is a curated collection of texts organized by difficulty, designed to help learners move systematically from simple to complex language structures.

This page outlines the rationale, design, and potential uses of such a corpus, aiming to provide educators, researchers, and technologists with a clear picture of its structure and benefits.

Why a Corpus Is Needed

• Scarcity of graded materials: Most Arabic textbooks rely on ungraded literary excerpts, making it hard to match texts to learners' proficiency.
• Standardized assessment: A leveled corpus offers a benchmark for measuring reading growth.
• Technology development: Machinelearning models for readability prediction, automated tutoring, and adaptive learning require large, annotated datasets.
• Linguistic research: Researchers can study lexical frequency, syntactic complexity, and discourse features across proficiency levels.

Design Principles

To be useful across contexts, the corpus follows four guiding principles:

1. Representativeness: Texts should cover a wide range of genres news, narrative, scientific exposition, and cultural essays reflecting everyday written MSA.
2. Transparency of levels: Each level must be defined using quantifiable readability metrics (e.g., word length, typetoken ratio, average sentence length) and validated through expert judgment.
3. Rich annotation: Morphological, syntactic, and discourse tags enable finegrained analysis and support NLP tasks.
4. Open access: The dataset will be released under a permissive license, encouraging reuse while respecting copyright.

Reading Levels

The corpus is organized into six levels (A1C2) loosely aligned with the Common European Framework of Reference for Languages (CEFR). The following table summarizes the main linguistic criteria for each level.

Content Types

Each level contains a balanced mix of the following genres:

• News articles: Short reports from reputable Arabic press, simplified for lower levels.
• Childrens stories: Traditional folk tales and modern narratives with illustrations (text only in the corpus).
• Scientific texts: Excerpts from textbooks on biology, geography, and basic physics.
• Cultural essays: Articles on art, history, and daily life customs.
• Dialogues: Simulated conversations that showcase everyday pragmatics.

All texts have been cleared for copyright or are drawn from publicdomain sources.

Annotation Scheme

Annotations are performed using the following layers:

1. Morphology: Tokenization, lemma, partofspeech (POS) tags, and detailed feature set (gender, number, case, definiteness).
2. Syntax: Dependency parsing (Universal Dependencies format) providing headdependent relations.
3. Discourse: Connective identification, rhetorical structure (Rhetorical Structure Theory), and cohesion markers.
4. Readability scores: Calculated via a modified Arabic Readability Index (ARI) and a machinelearned difficulty classifier.

Annotations are stored in standardized CoNLLU files, making the corpus directly usable with most NLP toolkits.

Applications

The corpus supports a wide range of educational and research activities:

• Curriculum design: Teachers can select texts that match lesson objectives and learners proficiency.
• Automated assessment: Readability models trained on the corpus can grade student essays or suggest appropriate reading material.
• Computerassisted language learning (CALL): Adaptive reading platforms can draw from the corpus to present texts that gradually increase in difficulty.
• Linguistic research: Studies on lexical development, syntactic change, or discourse strategies across proficiency levels.
• Speech synthesis & TTS: Levelspecific texts help train voice models that speak at a suitable pace and complexity for learners.

Access and Use

The corpus is hosted on a public GitHub repository and a permanent archive on Zenodo. Users can download:

• Raw text files sorted by level and genre.
• Annotated files (CoNLLU) for each layer.
• Metadata sheets containing source information, licensing, and readability statistics.

To encourage community contributions, a set of guidelines for adding new texts and annotations is provided. Contributions are reviewed by a panel of Arabic language specialists before integration.

Future Directions

Planned enhancements include:

• Multimodal extensions: Alignment of audio recordings with the written texts for listeningcomprehension practice.
• Dynamic difficulty modeling: Incorporating learner interaction data to refine difficulty scores in real time.
• Crossdialect comparison: Adding parallel texts in major Arabic dialects for contrastive studies.
• Pedagogical tools: Developing browserbased annotation interfaces that allow teachers to create custom leveled bundles.

Through these expansions, the corpus aims to become a central resource for the Arabic language learning ecosystem worldwide.