Sentence Level Quality Estimation on JFLEG (test)
61.61GLEUELECTRA-VERNet
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ELECTRA-VERNetbackbone=ELECTRA2021.05 | 61.61 | 38.63 | |
| BERT-VERNetbackbone=BERT2021.05 | 61.36 | 36.98 | |
| BERT-GED (JOINT)backbone=BERT, input=Joint Source + Hypothesis2021.05 | 60.49 | 33.03 | |
| BERT-GED (SRC)backbone=BERT, input=Source only2021.05 | 60.32 | 27.28 | |
| BERT-GQEbackbone=BERT2021.05 | 60.17 | 29.05 | |
| BERT-GED (HYP)backbone=BERT, input=Hypothesis2021.05 | 60 | 29.9 | |
| BERT-QEbackbone=BERT2021.05 | 59.73 | 26.16 | |
| BERT-LMbackbone=BERT2021.05 | 59.72 | 26.85 | |
| NQE (CC)variant=CC2021.05 | 57.22 | 7.68 | |
| NQE (CR)variant=CR2021.05 | 56.92 | 6.48 | |
| NQE (RC)variant=RC2021.05 | 56.91 | 1.66 | |
| NQE (RR)variant=RR2021.05 | 55.9 | 1.29 |