Contextual Machine Translation on anaphora benchmark
0.54BLEUgpt-4o
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| gpt-4oreasoning=with2025.10 | 0.54 | 0.71 | 0.92 | 0.91 | |
| gpt-4reasoning=with2025.10 | 0.53 | 0.71 | 0.93 | 0.92 | |
| gpt-4reasoning=without2025.10 | 0.49 | 0.7 | 0.92 | 0.92 | |
| gpt-4oreasoning=without2025.10 | 0.49 | 0.7 | 0.92 | 0.91 | |
| gpt-4-turboreasoning=with2025.10 | 0.49 | 0.67 | 0.9 | 0.9 | |
| gpt-3.5-turboreasoning=with2025.10 | 0.49 | 0.68 | 0.91 | 0.9 | |
| Phi-4reasoning=with2025.10 | 0.49 | 0.68 | 0.92 | 0.91 | |
| Llama 3.3reasoning=with2025.10 | 0.47 | 0.67 | 0.91 | 0.9 | |
| gpt-4-turboreasoning=without2025.10 | 0.45 | 0.67 | 0.92 | 0.91 | |
| gpt-3.5-turboreasoning=without2025.10 | 0.44 | 0.66 | 0.91 | 0.91 | |
| Llama 3.3reasoning=without2025.10 | 0.44 | 0.66 | 0.92 | 0.9 | |
| Phi-4reasoning=without2025.10 | 0.43 | 0.64 | 0.91 | 0.88 | |
| DeepSeek-R1 32Breasoning=with2025.10 | 0.39 | 0.59 | 0.84 | 0.84 | |
| nllb-200reasoning=without2025.10 | 0.37 | 0.6 | 0.9 | 0.87 | |
| DeepSeek-R1 32Breasoning=without2025.10 | 0.35 | 0.59 | 0.89 | 0.87 | |
| Llama 3.1reasoning=without2025.10 | 0.34 | 0.58 | 0.89 | 0.86 | |
| DeepSeek-R1 14Breasoning=without2025.10 | 0.34 | 0.58 | 0.89 | 0.86 | |
| DeepSeek-R1 14Breasoning=with2025.10 | 0.33 | 0.53 | 0.79 | 0.79 | |
| Llama 3.1reasoning=with2025.10 | 0.3 | 0.54 | 0.87 | 0.85 | |
| Mistralreasoning=without2025.10 | 0.27 | 0.51 | 0.86 | 0.82 | |
| Mistralreasoning=with2025.10 | 0.27 | 0.5 | 0.85 | 0.8 | |
| Llama 3.2reasoning=without2025.10 | 0.27 | 0.51 | 0.87 | 0.82 | |
| DeepSeek-R1 8Breasoning=without2025.10 | 0.24 | 0.5 | 0.86 | 0.8 | |
| Llama 3.2reasoning=with2025.10 | 0.23 | 0.47 | 0.83 | 0.76 | |
| DeepSeek-R1 8Breasoning=with2025.10 | 0.21 | 0.45 | 0.77 | 0.77 |