Question Answering on MultiSpanQA (test)
71.4Exact F1LIQUID + RoBERTalarge + Tagger
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| LIQUID + RoBERTalarge + TaggerTraining Setup=synthetic & labeled (D' -> D), Backbone=RoBERTa-large, Model Type=Sequence tagging2023.02 | 71.4 | 75 | 68.2 | 80.9 | 85.3 | 77 | |
| LIQUID + RoBERTabase + TaggerTraining Setup=synthetic & labeled (D' -> D), Backbone=RoBERTa-base, Model Type=Sequence tagging2023.02 | 67.4 | 65.7 | 69.2 | 81.2 | 80.9 | 81.5 | |
| RoBERTalarge + TaggerTraining Setup=labeled only (D), Backbone=RoBERTa-large, Model Type=Sequence tagging2023.02 | 66.4 | 62.3 | 71.2 | 82.6 | 82.1 | 83 | |
| LIQUID + CorefRoBERTalarge + TaggerTraining Setup=synthetic & labeled (D' -> D), Backbone=CorefRoBERTa-large, Model Type=Sequence tagging2023.02 | 65.8 | 64 | 67.8 | 80.2 | 79.8 | 80.5 | |
| CorefRoBERTalarge + TaggerTraining Setup=labeled only (D), Backbone=CorefRoBERTa-large, Model Type=Sequence tagging2023.02 | 64 | 56.5 | 73.8 | 81.7 | 77.7 | 86 | |
| RoBERTabase + TaggerTraining Setup=labeled only (D), Backbone=RoBERTa-base, Model Type=Sequence tagging2023.02 | 62.9 | 63 | 62.9 | 78 | 82.5 | 73.9 | |
| BERTbase + Tagger (multi-task)Training Setup=labeled only (D), Backbone=BERT-base, Model Type=Multi-task sequence tagging2023.02 | 59.3 | 58.1 | 60.5 | 76.3 | 79.6 | 73.2 | |
| BERTbase + TaggerTraining Setup=labeled only (D), Backbone=BERT-base, Model Type=Sequence tagging2023.02 | 56.5 | 52.5 | 61.1 | 75.2 | 75.9 | 74.5 | |
| LIQUID + RoBERTabase + Single-spanTraining Setup=synthetic & labeled (D' -> D), Backbone=RoBERTa-base, Model Type=Single-span extractive2023.02 | 19.4 | 19.7 | 19 | 71 | 62.9 | 81.4 | |
| BERTbase + Single-spanTraining Setup=labeled only (D), Backbone=BERT-base, Model Type=Single-span extractive2023.02 | 14.4 | 16.2 | 13 | 67.6 | 60.3 | 76.8 | |
| RoBERTabase + Single-spanTraining Setup=labeled only (D), Backbone=RoBERTa-base, Model Type=Single-span extractive2023.02 | 10.5 | 14.4 | 8.3 | 63 | 60 | 66.3 |