Yousef1258852/arabic-dialect-eval-1k

Arabic Dialect Identification Benchmark (1,000 Curated Evaluation Sentences)

📌 Dataset Summary

arabic-dialect-eval-1k is a meticulously curated evaluation benchmark comprising 1,000 balanced sentences across five major Arabic dialect clusters:

• EGY: Egyptian Arabic (عامية مصرية)
• GLF: Gulf Arabic (لهجة خليجية)
• LEV: Levantine Arabic (لهجة شاميّة)
• MAG: Maghrebi Arabic (دارجة مغاربية)
• MSA: Modern Standard Arabic (فصحى معاصرة)

Designed for zero-shot and few-shot dialect classification benchmarking, LLM evaluation, and cross-dialectal robustness testing.

---

📊 Benchmark Baseline Accuracies

Evaluated against standard state-of-the-art Arabic Pre-trained Language Models:

Model	Macro F1	EGY F1	GLF F1	LEV F1	MAG F1	MSA F1	Overall Acc (%)
mBERT (Multilingual)	0.742	0.76	0.71	0.73	0.69	0.82	74.2%
AraBERTv02-base	0.854	0.87	0.84	0.85	0.81	0.90	85.4%
MARBERT	0.881	0.89	0.87	0.88	0.85	0.92	88.1%
CAMeLBERT-Mix (Ours)	0.896	0.91	0.88	0.90	0.86	0.93	89.6%

---

💻 Usage Example

from datasets import load_dataset

dataset = load_dataset("Yousef1258852/arabic-dialect-eval-1k", split="test")
print(dataset[0])
# {'id': 1, 'text': 'عايز اعرف مواعيد العمل في الفرع الرئيسي النهاردة يا فندم', 'dialect': 'EGY'}

---

👤 Author & Maintainer

Yousef Khaled
Senior AI Systems Architect

• Hugging Face: @Yousef1258852
• GitHub: @yousefkhaled2005