Machine Reading Comprehension on RACE
94.5RACE Overall AccuracyHuman Ceiling Performance
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Human Ceiling PerformancePre-training=None2019.03 | 94.5 | 95.4 | 94.2 | |
| OCN_LARGEPre-training=BERT, Configuration=Ensemble, Backbone=BERT_LARGE2019.03 | 73.5 | 78.4 | 71.5 | |
| Amazon Mechanical TurkerPre-training=None2019.03 | 73.3 | 85.1 | 69.4 | |
| DCMNPre-training=BERT, Configuration=Ensemble2019.03 | 72.1 | 77.9 | 69.8 | |
| OCN_LARGEPre-training=BERT, Configuration=Single Model, Backbone=BERT_LARGE2019.03 | 71.7 | 76.7 | 69.6 | |
| OursBackbone=BERT-base2022.03 | 71.01 | — | — | |
| BERT_LARGEPre-training=BERT, Configuration=Single Model2019.03 | 71 | 76.4 | 68.8 | |
| DCMNPre-training=BERT, Configuration=Single Model2019.03 | 70.9 | 76.7 | 68.5 | |
| OCN_BASEPre-training=BERT, Configuration=Ensemble, Backbone=BERT_BASE2019.03 | 69.2 | 74.4 | 67 | |
| B+C+VcorrBackbone=BERT-base2022.03 | 68.66 | — | — | |
| B+CBCLBackbone=BERT-base2022.03 | 67.79 | — | — | |
| B+DRCABackbone=BERT-base2022.03 | 67.48 | — | — | |
| B+C+PredictabilityBackbone=BERT-base2022.03 | 66.82 | — | — | |
| OCN_BASEPre-training=BERT, Configuration=Single Model, Backbone=BERT_BASE2019.03 | 66.8 | 71.6 | 64.8 | |
| Reading Strategies ModelPre-training=BERT, Configuration=Ensemble2019.03 | 66.7 | 72 | 64.5 | |
| B+CL+VBackbone=BERT-base2022.03 | 66.68 | — | — | |
| B+C+v3Backbone=BERT-base2022.03 | 65.82 | — | — | |
| B+C+v4Backbone=BERT-base2022.03 | 65.73 | — | — | |
| BERT_BASEPre-training=BERT, Configuration=Single Model2019.03 | 65.2 | 70.5 | 63 | |
| B+C+ForgettingBackbone=BERT-base2022.03 | 64.91 | — | — | |
| B+C+v2Backbone=BERT-base2022.03 | 64.11 | — | — | |
| B+C+v1Backbone=BERT-base2022.03 | 63.92 | — | — | |
| Reading Strategies ModelPre-training=GPT, Configuration=Single Model2019.03 | 63.8 | 69.2 | 61.5 | |
| B+C+DatasetMapBackbone=BERT-base2022.03 | 63.51 | — | — | |
| BBackbone=BERT-base2022.03 | 63.02 | — | — | |
| B+antiCL+VBackbone=BERT-base2022.03 | 61.8 | — | — | |
| OpenAI GPTPre-training=GPT, Configuration=Single Model2019.03 | 59 | 62.9 | 57.4 | |
| MRUPre-training=None, Configuration=Ensemble2019.03 | 53.3 | 60.2 | 50.3 | |
| DFNPre-training=None, Configuration=Ensemble2019.03 | 51.2 | 55.6 | 49.4 | |
| Hier-Co-MatchingPre-training=None, Configuration=Single Model2019.03 | 50.4 | 55.8 | 48.2 | |
| MRUPre-training=None, Configuration=Single Model2019.03 | 50.4 | 57.7 | 47.4 | |
| DFNPre-training=None, Configuration=Single Model2019.03 | 47.4 | 51.5 | 45.7 | |
| ElimiNetPre-training=None, Configuration=Ensemble2019.03 | 46.5 | 47.7 | 46.1 | |
| HAFPre-training=None, Configuration=Single Model2019.03 | 46 | 45 | 46.4 | |
| GA ReaderPre-training=None, Configuration=Ensemble2019.03 | 45.9 | — | — | |
| ElimiNetPre-training=None, Configuration=Single Model2019.03 | 44.5 | 44.4 | 44.5 | |
| GA ReaderPre-training=None, Configuration=Single Model2019.03 | 44.1 | 43.7 | 44.2 | |
| Stanford ARPre-training=None, Configuration=Single Model2019.03 | 43.3 | 44.2 | 43 |