Token Level Quality Estimation on CoNLL ann. 2 2014 (test)
82.25PrecisionBERT-VERNet
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| BERT-VERNetScenario=Source, Input Encoding=Source, hypothesis pairs2021.05 | 82.25 | 28.49 | 59.71 | |
| ELECTRA-VERNetScenario=Source, Input Encoding=Source, hypothesis pairs2021.05 | 81.69 | 32.97 | 63.06 | |
| BERT-GED (SRC)Scenario=Source, Input Encoding=Source sentences only2021.05 | 77.94 | 25.02 | 54.77 | |
| BERT-GED (JOINT)Scenario=Source, Input Encoding=Source, hypothesis pairs2021.05 | 77.42 | 25.23 | 54.77 | |
| ELECTRA-VERNetScenario=Hypothesis, Input Encoding=Source, hypothesis pairs2021.05 | 72.55 | 34.42 | 59.39 | |
| BERT-VERNetScenario=Hypothesis, Input Encoding=Source, hypothesis pairs2021.05 | 71.79 | 29.04 | 55.46 | |
| BERT-GED (HYP)Scenario=Hypothesis, Input Encoding=Source, hypothesis pairs2021.05 | 66.49 | 27.68 | 51.93 | |
| BERT-GED (JOINT)Scenario=Hypothesis, Input Encoding=Source, hypothesis pairs2021.05 | 64.79 | 31.52 | 53.5 |