Multilingual Summarization on XL-SUM ar
0.37Token-level Language Confusion (%)LATB
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LATBBackbone=Llama3 8B-I2026.06 | 0.37 | 0.28 | 0 | |
| Adaptive LATBBackbone=Llama3 8B-I2026.06 | 0.54 | 0.22 | 0 | |
| Suzume 8B-Multilingual2026.06 | 5.63 | 2.95 | 2.6 | |
| Llama3 8B-IPrompting Strategy=Strict Prompt2026.06 | 9.98 | 11.94 | 5.6 | |
| Llama3 8B-IPrompting Strategy=Standard2026.06 | 86.21 | 99.27 | 98.3 |