| # Data Sources and Composition |
|
|
| The dataset contains **18,935,335 directional translation examples** across 21 source pools. Percentages below are calculated from the final example counts, not token counts or unique bilingual pairs. |
|
|
| | Source | Examples | Share | |
| |---|---:|---:| |
| | UNPC | 3,704,842 | 19.566% | |
| | OpenSubtitles | 2,708,074 | 14.302% | |
| | MultiUN | 2,377,741 | 12.557% | |
| | CCMatrix | 1,580,197 | 8.345% | |
| | OPUS100 | 1,554,605 | 8.210% | |
| | WikiMatrix | 1,415,377 | 7.475% | |
| | CCAligned — additional OPUS pool (`opus_big_ccaligned`) | 1,080,624 | 5.707% | |
| | CCAligned (`ccaligned`) | 955,073 | 5.044% | |
| | News Commentary | 951,488 | 5.025% | |
| | CCMatrix — additional OPUS pool (`opus_big_ccmatrix`) | 764,627 | 4.038% | |
| | BWB | 486,810 | 2.571% | |
| | Wikimedia | 396,975 | 2.096% | |
| | TED2013 | 360,985 | 1.906% | |
| | WikiTitles | 293,505 | 1.550% | |
| | CoVoST2 | 232,457 | 1.228% | |
| | BSTC | 27,788 | 0.147% | |
| | TED2020 | 17,934 | 0.095% | |
| | QED | 13,955 | 0.074% | |
| | MASSIVE | 8,644 | 0.046% | |
| | Neulab TED Talks | 3,604 | 0.019% | |
| | ELRC Health | 30 | <0.001% | |
| | **Total** | **18,935,335** | **100%** | |
|
|
| ## Translation Directions |
|
|
| | Direction | Examples | Share | |
| |---|---:|---:| |
| | English to Chinese | 10,106,394 | 53.373% | |
| | Chinese to English | 8,828,941 | 46.627% | |
|
|
| `src` identifies the source pool for each record. The additional OPUS pools are listed separately to match those labels. Percentages are rounded and may not sum to exactly 100%. |
|
|
