Update README.md

README.md

@@ -20,9 +20,29 @@ This Space provides an interactive viewer for the **300K English-Vietnamese-Japa
 
 - **Languages:** English, Vietnamese, Japanese
 - **Size:** 300,000 sentence-level aligned triplets
-- **Source:** Curated from OPUS parallel corpora (CCMatrix, OpenSubtitles, JParaCrawl, WikiMatrix)
 - **Format:** JSONL with `en`, `vi`, `ja` keys
 
+## Data Sources
+
+This dataset is curated from [OPUS](https://opus.nlpl.eu/) parallel corpora:
+
+| Corpus | Language Pair | Description |
+|--------|---------------|-------------|
+| CCMatrix v1 | EN-VI, EN-JA | Web-crawled parallel sentences |
+| OpenSubtitles v2024 | EN-VI | Movie and TV subtitle translations |
+| JParaCrawl v3.0 | EN-JA | Japanese web parallel corpus |
+| WikiMatrix v1 | EN-JA | Wikipedia-extracted translations |
+
+## Data Pipeline
+
+1. **Collection** - Downloaded parallel sentence pairs from OPUS
+2. **Cleaning** - Removed noise, special characters, and malformed text
+3. **Filtering** - Kept sentences with appropriate length and quality
+4. **Language Verification** - Validated correct languages in each pair
+5. **Deduplication** - Removed duplicate and near-duplicate sentences
+6. **Triplet Creation** - Joined EN-VI and EN-JA pairs using English as pivot
+7. **Final Sampling** - Selected 300K high-quality triplets
+
 ## Usage
 
 Click the "Sample random triplet" button to view random examples from the dataset.