Machine Translation on Reference-based Mongolian-to-English Translation Dataset
14.16BLEU-3 ScoreCoPiT
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| CoPiTBackbone=Ministral-3 (14B), Evaluation Protocol=Fine-tuned2026.07 | 14.16 | 9.58 | 38.24 | 35.41 | 70.7 | |
| CoPiTBackbone=GPT-4.1, Evaluation Protocol=Zero-shot2026.07 | 14.07 | 9.83 | 37.29 | 24.59 | 66.3 | |
| CoPiTBackbone=Qwen-3 (4B), Evaluation Protocol=Fine-tuned2026.07 | 9.4 | 5.7 | 29.7 | 27.55 | 62.8 | |
| CoPiTBackbone=Qwen-3 (30B), Evaluation Protocol=Fine-tuned2026.07 | 7.61 | 4.56 | 29.25 | 26.94 | 59.3 | |
| CoPiTBackbone=Ministral-3 (3B), Evaluation Protocol=Fine-tuned2026.07 | 4.81 | 2.66 | 29.38 | 26.21 | 61.9 | |
| CoPiTBackbone=Ministral-3 (14B), Evaluation Protocol=Zero-shot2026.07 | 3.02 | 1.4 | 26.02 | 23.3 | 54.9 | |
| DirectBackbone=GPT-4.1, Evaluation Protocol=Direct2026.07 | 2.75 | 1.35 | 19.59 | 17.66 | 46.8 | |
| CoPiTBackbone=Qwen-3 (30B), Evaluation Protocol=Zero-shot2026.07 | 1.86 | 0.92 | 21.75 | 19.28 | 49.3 | |
| DirectBackbone=Ministral-3 (14B), Evaluation Protocol=Direct2026.07 | 1.79 | 0.77 | 21.44 | 19.13 | 46.7 | |
| DirectBackbone=Qwen-3 (30B), Evaluation Protocol=Direct2026.07 | 1.25 | 0.48 | 19.73 | 17.51 | 44.3 | |
| DirectBackbone=Ministral-3 (3B), Evaluation Protocol=Direct2026.07 | 1.21 | 0.46 | 23.19 | 20.21 | 50 | |
| CoPiTBackbone=Ministral-3 (3B), Evaluation Protocol=Zero-shot2026.07 | 1.09 | 0.22 | 22.66 | 19.82 | 49.7 | |
| CoPiTBackbone=Qwen-3 (4B), Evaluation Protocol=Zero-shot2026.07 | 0.94 | 0.46 | 15.57 | 13.73 | 42.2 | |
| DirectBackbone=Qwen-3 (4B), Evaluation Protocol=Direct2026.07 | 0.54 | 0.21 | 13.34 | 11.73 | 37.3 |