Grounded Commonsense Inference on SWAG (test)
88AccuracyHuman (5 annotations)
Evaluation Results
| Method | Links | |
|---|---|---|
| Human (5 annotations)sample size=1002018.10 | 88 | |
| BERT_LARGEepochs=3, learning rate=2e-5, batch size=162018.10 | 86.3 | |
| Human (expert)sample size=1002018.10 | 85 | |
| OpenAI GPT2018.10 | 78 | |
| ESIM+ELMo2018.10 | 59.2 | |
| ESIM+GloVe2018.10 | 52.7 |