Add Egyptian Arabic <-> English translation dataset (50K) + card

README.md

@@ -0,0 +1,116 @@
+---
+license: apache-2.0
+task_categories:
+  - translation
+language:
+  - ar
+  - en
+tags:
+  - egyptian-arabic
+  - arabic
+  - egyptian
+  - dialectal-arabic
+  - translation
+  - machine-translation
+  - parallel-corpus
+size_categories:
+  - 10K<n<100K
+pretty_name: Egyptian Arabic ↔ English Translation (50K)
+configs:
+  - config_name: default
+    data_files:
+      - split: train
+        path: data/train-*.parquet
+---
+
+# Egyptian Arabic ↔ English Translation (50K)
+
+An **open-source** parallel corpus of **50,000** Egyptian Arabic sentences paired
+with their English translations. Released to help the community build and evaluate
+translation systems for **Egyptian / dialectal Arabic** — a widely spoken variety
+that remains under-served by open translation data compared to Modern Standard
+Arabic.
+
+Use it freely for machine translation, LLM fine-tuning and evaluation, dialectal
+Arabic NLP, and data augmentation.
+
+## Dataset structure
+
+A single `train` split with 50,000 rows and two text columns:
+
+| Column | Language | Description |
+|---|---|---|
+| `Arabic` | Egyptian Arabic (`ar`) | A sentence in colloquial Egyptian Arabic |
+| `English` | English (`en`) | Its English translation |
+
+The pairs are aligned one-to-one, so the dataset can be used in **either
+direction** (Egyptian Arabic → English or English → Egyptian Arabic).
+
+### Examples
+
+| Arabic | English |
+|---|---|
+| تعالى شوفها | Come see it |
+| دي فكرة ممتازة | That's a great idea |
+| اتأكد إن مفيش كاميرات. | Make sure there are no cameras. |
+| شكراً يا أستاذ. | Thank you, dear professor. |
+| عمرك رحت زفتى في التمانينات؟ | Have you ever been to Zefta in the 80s? |
+
+## Usage
+
+```python
+from datasets import load_dataset
+
+ds = load_dataset("oddadmix/egyptian-arabic-english-translation-50k", split="train")
+print(ds[0])
+# {'Arabic': '...', 'English': '...'}
+
+# Build a supervised translation pair
+for ex in ds.select(range(3)):
+    src, tgt = ex["Arabic"], ex["English"]
+```
+
+Or straight from the parquet with pandas / polars:
+
+```python
+import pandas as pd
+df = pd.read_parquet("data/train-00000-of-00001.parquet")
+```
+
+## Intended uses
+
+- Training and fine-tuning **Egyptian Arabic ↔ English** translation models.
+- Fine-tuning or instruction-tuning LLMs for dialectal Arabic understanding and generation.
+- Building evaluation / benchmark sets for dialectal machine translation.
+- Data augmentation for low-resource Arabic dialect NLP.
+
+## Notes and limitations
+
+- The Arabic side is **colloquial Egyptian**, so spelling and phrasing vary
+  naturally (dialectal Arabic has no single standardized orthography). Expect
+  informal register, code-switching, and conversational style.
+- Sentence lengths range from a few words to full paragraphs.
+- No predefined train/validation/test split is provided — create your own split
+  to fit your setup (e.g. `ds.train_test_split(test_size=0.05, seed=42)`).
+- As with any translation corpus, occasional imperfect alignments may exist;
+  filter or clean further if your task is sensitive to noise.
+
+## License
+
+Released under the **Apache-2.0** license — free to use, modify, and redistribute,
+including commercially, with attribution. If Apache-2.0 doesn't fit your needs,
+open a discussion on the dataset page.
+
+## Citation
+
+```bibtex
+@misc{egyptian_arabic_english_translation_50k,
+  title  = {Egyptian Arabic--English Translation (50K)},
+  author = {oddadmix},
+  year   = {2026},
+  url    = {https://huggingface.co/datasets/oddadmix/egyptian-arabic-english-translation-50k}
+}
+```
+
+Contributions, corrections, and issues are welcome — this dataset is shared to
+support open work on Egyptian Arabic. 🇪🇬

data/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac25d3ceff020650b51cb9ad02e79d4ba98f81ff7185849364a4ae2cc916683e
+size 12536940