Summarization on XSum (ROUGE-L, COMET, P/R/F1)
21.6ROUGE-LSeqMark
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| SeqMark2026.01 | 21.6 | 68.5 | 81.3 | 100 | 89.7 | 85.3 | 85.3 | 85.3 | — | |
| k-SemStamp2026.01 | 20.7 | 68.9 | 54.2 | 22 | 31.3 | 61.6 | 22 | 32.4 | — | |
| No Watermark2026.01 | 20.5 | 69 | — | — | — | — | — | — | — | |
| KGW2026.01 | 20.1 | 68.8 | 85.2 | 30.1 | 44.9 | 79.8 | 30.4 | 44 | — | |
| SemStamp2026.01 | 20 | 68.7 | 70.6 | 52.8 | 60.4 | 67.8 | 59 | 63.1 | — | |
| SWEET2026.01 | 18.5 | 68.7 | 52.4 | 49.9 | 51.1 | 53 | 40 | 45.6 | — | |
| SummaCYear=2022, Key Approach=NLI-based consistency2026.06 | — | — | — | — | — | — | — | — | 84.5 |