Task-oriented Dialogue on FewShotSGD seen schemata (test)
29.28BLEUFull Data Baseline
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Full Data BaselineTrain split=164,9782021.10 | 29.28 | 1.12 | |
| BLEURT self-trainingPseudo-response selection strategy=BLEURT, Train split=40-shot (4,312)2021.10 | 27.48 | 2.37 | |
| Vanilla self-trainingPseudo-response selection strategy=Vanilla, Train split=40-shot (4,312)2021.10 | 26.65 | 5 | |
| BLEURT self-trainingPseudo-response selection strategy=BLEURT, Train split=20-shot (2,140)2021.10 | 26.63 | 3.33 | |
| T5-smallPseudo-response selection strategy=None, Train split=40-shot (4,312)2021.10 | 25.72 | 7.6 | |
| BLEURT self-trainingPseudo-response selection strategy=BLEURT, Train split=10-shot (1,075)2021.10 | 25.63 | 4.29 | |
| BLEURT self-trainingPseudo-response selection strategy=BLEURT, Train split=5-shot (558)2021.10 | 25.22 | 4.78 | |
| Vanilla self-trainingPseudo-response selection strategy=Vanilla, Train split=10-shot (1,075)2021.10 | 23.5 | 17.9 | |
| Vanilla self-trainingPseudo-response selection strategy=Vanilla, Train split=20-shot (2,140)2021.10 | 23.19 | 14.92 | |
| Vanilla self-trainingPseudo-response selection strategy=Vanilla, Train split=5-shot (558)2021.10 | 23.03 | 15.15 | |
| T5-smallPseudo-response selection strategy=None, Train split=20-shot (2,140)2021.10 | 22.84 | 16.74 | |
| T5-smallPseudo-response selection strategy=None, Train split=10-shot (1,075)2021.10 | 21.45 | 21.64 | |
| T5-smallPseudo-response selection strategy=None, Train split=5-shot (558)2021.10 | 20.66 | 22.84 |