Table 6:
SARI and error distributions in system outputs manually evaluated.
| Insertion | Deletion | Substitution | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | Dataset | SARI | 0 | 1 | 2 | −1 | 0 | 1 | 2 | −1 | 0 | 1 | 2 | −1 |
| Dress | Wikilarge | 34.9 | 91.9 | 0.8 | 0.8 | 6.5 | 42.6 | 24.6 | 26.2 | 6.6 | 84.4 | 4.1 | 4.9 | 6.6 |
| Newsela | 34.5 | 90.5 | 0.0 | 0.0 | 9.5 | 29.9 | 29.2 | 32.1 | 9.7 | 67.4 | 6.5 | 15.9 | 10.1 | |
| EditNTS | Wikilarge | 40.4 | 94.3 | 4.9 | 0.8 | 0.0 | 55.0 | 24.2 | 20.8 | 0.0 | 88.5 | 4.1 | 7.4 | 0.0 |
| Newsela | 36.3 | 69.4 | 0.7 | 2.7 | 27.2 | 9.5 | 19.0 | 44.2 | 27.2 | 64.4 | 2.1 | 6.2 | 27.4 | |
| T5 | Wikilarge | 34.9 | 96.8 | 1.6 | 0.8 | 0.8 | 81.6 | 14.4 | 3.2 | 0.8 | 97.6 | 1.6 | 0.0 | 0.8 |
| Newsela | 38.6 | 81.7 | 9.6 | 7.0 | 1.7 | 27.7 | 43.7 | 26.9 | 1.7 | 92.4 | 5.9 | 0.0 | 1.7 | |
| Access | Wikilarge | 49.7 | 89.1 | 8.2 | 0.9 | 1.8 | 57.5 | 34.9 | 5.7 | 1.9 | 71.1 | 18.6 | 8.2 | 2.1 |
| ControlTS | Wikilarge | 42.3 | 88.8 | 7.8 | 1.7 | 1.7 | 47.8 | 39.1 | 11.3 | 1.7 | 81.5 | 15.1 | 1.7 | 1.7 |