N1K0L0Z/N1K0_ka_en_corpora

N1K0 Georgian-English Corpora

A high-quality, heavily sanitized parallel translation dataset containing ~2 Million (1,988,071) parallel sentence pairs for Georgian (ka) and English (en).

This dataset is specifically engineered for neural machine translation tasks, sequence-to-sequence models (e.g., Transformers), and cross-attention decoder pretraining/fine-tuning.

---

🚀 Quick Start & Usage

You can load this dataset directly using the Hugging Face datasets library:

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("N1K0L0Z/N1K0_ka_en_corpora")
train_data = dataset["train"]

print(f"Total parallel sentences: {len(train_data):,}")
print("Sample Pair:", train_data[0])

---

📊 Dataset Statistics

Metric	Georgian (KA)	English (EN)	Total / Combined
Total Parallel Sentences	1,988,071	1,988,071	1,988,071
Total Word Count	22,810,906	29,718,347	52,529,253
Total Characters	209,617,205	207,073,997	416,691,202
Average Words / Sentence	11.47	14.95	—
Average Chars / Sentence	105.44	104.16	—
Min / Max Words	2 / 46	2 / 60	—
Unique Vocabulary Size	210,276	60,833	271,109

---

🛠 Generation & Targeted Entity Enrichment

This dataset was curated and translated leveraging Google's Gemini 2.5 Flash model.

To prevent common neural machine translation failure modes (such as hallucinating or dropping proper nouns, numbers, and rare entities), the data synthesis pipeline explicitly injected localized, high-value entity domains across both Georgian and English:

• 👤 Bilingual Names & Surnames: Complete coverage of authentic Georgian and English first names and family names to ensure proper transliteration and entity retention.
• 📍 Geographic Landmarks & Addresses: Comprehensive regional naming conventions, street addresses, and geographic locations across Georgia and international destinations.
• 🔢 Complex Numerics & Expressions: Varied numeric formats, dates, monetary units, and mathematical phrasing in both languages.
• 🐾 Fauna & Nature Vocabulary: Granular species names, animal terminology, and natural environment phrasing.
• 💬 Idioms & Figurative Language: Natural colloquialisms, expressions, and cultural idioms native to both languages to prevent stiff, literal translation artifacts.

---

🧹 Data Cleaning & Filtering Pipeline

To ensure high data quality, the raw output passed through a strict 7-stage cleaning pipeline:

• 1. Deduplication: Removed all identical sentence pairs.
• 2. Short Entry Removal: Dropped isolated 1-word entries and incomplete fragments.
• 3. Symbol Cleaning: Removed unwanted formatting symbols (*, ^, {}, []) and fixed irregular spacing.
• 4. Bracket Alignment:
  • Removed short noise in brackets (like 2 words or fewer, e.g., "(e.g.)" or "(1)").
  • Dropped pairs where brackets appeared in one language but were missing in the other.
• 5. Georgian Hyphen Fixing: Standardized attached grammatical suffix hyphens (such as -ში, -ს, -ის).
• 6. Repetition & Artifact Guardrails: Dropped spaced-out characters (l i k e t h i s) and repeated word loops.
• 7. Length Ratio Filtering: Removed pairs where one sentence was unnaturally longer than the other.

---

📄 License & Citation

This dataset is released under the MIT License.

If you use this dataset in your research or translation pipelines, please cite:

@misc{nikoloz_ka_en_2026,
  author = {Nikoloz Kordzakhia},
  title = {N1K0 Georgian-English Corpora},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face Repository},
  howpublished = {\url{[https://huggingface.co/datasets/N1K0L0Z/N1K0_ka_en_corpora](https://huggingface.co/datasets/N1K0L0Z/N1K0_ka_en_corpora)}}
}