Multilingual Summarization on XL-SUM ru
0.28Token-level Language ConfusionLATB
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LATBBackbone=Llama3 8B-I2026.06 | 0.28 | 0.44 | 0.1 | |
| Adaptive LATBBackbone=Llama3 8B-I2026.06 | 0.48 | 0.38 | 0.1 | |
| Suzume 8B-Multilingual2026.06 | 3.04 | 2.3 | 2.1 | |
| Llama3 8B-IPrompting Strategy=Strict Prompt2026.06 | 5.02 | 4.1 | 2.9 | |
| Llama3 8B-IPrompting Strategy=Standard2026.06 | 80.66 | 93.83 | 92.5 |