| --- |
| license: cc-by-nc-4.0 |
| language: |
| - ar |
| task_categories: |
| - text-generation |
| - text-classification |
| tags: |
| - arabic |
| - msa |
| - modern-standard-arabic |
| - news |
| - native-written |
| pretty_name: Arabic Modern Standard (MSA) Sample — native-written |
| size_categories: |
| - n<1K |
| configs: |
| - config_name: default |
| data_files: "4factors_msa_sample_50.csv" |
| extra_gated_heading: Request access to the 4FACTORS Arabic samples |
| extra_gated_description: >- |
| These samples are free for research and non-commercial evaluation. Tell us |
| briefly what you are building — it helps us understand what the field |
| actually needs. |
| extra_gated_fields: |
| Company or institution: text |
| "What are you building?": text |
| I agree to use this data for non-commercial purposes only: checkbox |
| extra_gated_button_content: Request access |
| --- |
| |
| # Arabic — Modern Standard Arabic (MSA) Sample |
|
|
| Native-written, human-verified Modern Standard Arabic. No scraping. No machine |
| translation. No synthetic generation. Every sentence written from scratch by a |
| first-language speaker in formal news / official-statement register, then reviewed |
| line by line against a written checklist and measured for structural diversity across |
| the whole set. |
|
|
| A **public demonstration sample** (50 items). Larger MSA datasets and other varieties |
| (Levantine, Egyptian, Gulf, Maghrebi) are delivered on commission with full IP transfer. |
|
|
| This is the Modern Standard Arabic companion to |
| [`4factors/arabic-palestinian-levantine-sample`](https://huggingface.co/datasets/4factors/arabic-palestinian-levantine-sample). |
|
|
| Project page: https://www.4factors.ch/arabic-nlp-training-data/ |
|
|
|
|
| ## Columns |
|
|
| | Field | Description | |
| |-----------------|----------------------------------------------------------| |
| | id | Stable item id (`msa-001` … `msa-050`) | |
| | text_ar | The Arabic sentence (MSA) | |
| | translation_en | English gloss, added by the 4FACTORS team | |
| | domain | Domain label (news, health, economy, science, …) | |
| | variety | Always `Modern Standard Arabic` | |
|
|
| 50 rows, single `train` split. |
|
|
| ## How this sample was made |
|
|
| 1. **Scoped brief** — formal register, explicit prohibitions on copying, translating |
| and AI generation, one worked example as the quality bar. |
| 2. **Native composition** — one original MSA sentence per prompted topic. |
| 3. **Independent review** — a second native speaker marks every line: pass, preferred |
| change, or mandatory change. Findings recorded per row. |
| 4. **Diversity measurement** — sentence-opening repetition, connective frequency and |
| length distribution counted across all 50 items and brought within threshold |
| (≤3 sentences opening with the same verb; ≥15 sentences under 12 words). |
| 5. **Rewrite & re-check** — the author revises; the reviewer verifies. Both versions of |
| the file are retained, so the correction history is auditable. |
| 6. **Normalisation & delivery** — orthography normalised, English glosses added. |
|
|
| ## Provenance & ethics |
|
|
| - 100% human-written. Zero AI generation. |
| - No real personal names, brands, or copyrighted source text. |
| - The contributor is compensated and has consented to publication. |
|
|
| ## License |
|
|
| **CC BY-NC 4.0** — free to download and evaluate for non-commercial use, with |
| attribution. For commercial licensing or commissioned datasets (full IP transfer, your |
| schema, English glosses included), contact 4FACTORS: info@4factors.ch — |
| https://www.4factors.ch |
|
|
| ## Archived deposit |
|
|
| This sample is part of an archived corpus (all three varieties, 150 items) published on |
| Zenodo together with the full provenance record — contributor brief, review passes, |
| correction logs and SHA-256 checksums: |
|
|
| **DOI: [10.5281/zenodo.21380048](https://doi.org/10.5281/zenodo.21380048)** |
|
|
| ## Citation |
|
|
| > Poljanc, A., & 4FACTORS GmbH (2026). 4FACTORS Arabic Native-Written Sample Corpus: |
| > Modern Standard Arabic, Palestinian Levantine and Egyptian (150 items) (Version 1.0) |
| > [Data set]. Zenodo. https://doi.org/10.5281/zenodo.21380048 |
|
|
