Multilingual Summarization on XL-SUM ko
3.45Token-level Language ConfusionLATB
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LATBBackbone=Llama3 8B-I2026.06 | 3.45 | 9.98 | 10.36 | |
| Adaptive LATBBackbone=Llama3 8B-I2026.06 | 4.56 | 10.12 | 11.45 | |
| Suzume 8B-Multilingual2026.06 | 8.28 | 11.74 | 12.36 | |
| Llama3 8B-IPrompting Strategy=Strict Prompt2026.06 | 16.72 | 30.79 | 27.27 | |
| Llama3 8B-IPrompting Strategy=Standard2026.06 | 85.46 | 99.6 | 99.63 |