Multilingual Summarization on XL-SUM zh
4.78Token-level Language Confusion (%)LATB
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LATBBackbone=Llama3 8B-I2026.06 | 4.78 | 0 | 0 | |
| Adaptive LATBBackbone=Llama3 8B-I2026.06 | 5.37 | 0.1 | 0 | |
| Suzume 8B-Multilingual2026.06 | 7.56 | 0.89 | 0.9 | |
| Llama3 8B-IPrompting Strategy=Strict Prompt2026.06 | 14.17 | 9.69 | 9.1 | |
| Llama3 8B-IPrompting Strategy=Standard2026.06 | 85.92 | 98.9 | 98.9 |