Text Summarization on Annotated English Gigaword standard (test)
39.81ROUGE-1REP(UNI)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| REP(UNI)Position=both2021.04 | 39.81 | 20.4 | 36.93 | — | — | |
| REP(SIM)Position=both2021.04 | 39.7 | 20.14 | 36.77 | — | — | |
| WDROPPosition=both2021.04 | 39.66 | 20.45 | 36.59 | — | — | |
| REP(SIM)+WDROPPosition=both2021.04 | 39.56 | 20.14 | 36.66 | — | — | |
| Qi et al. (2020)2021.04 | 39.51 | 20.42 | 36.69 | — | — | |
| REP(UNI)+WDROPPosition=both2021.04 | 39.36 | 20.13 | 36.62 | — | — | |
| w/o perturbation2021.04 | 39.2 | 19.84 | 36.21 | — | — | |
| REP(SS)Position=dec2021.04 | 39.2 | 20.04 | 36.27 | — | — | |
| Zhang et al. (2020)2021.04 | 39.12 | 19.86 | 36.24 | — | — | |
| Song et al. (2019)2021.04 | 38.73 | 19.71 | 35.96 | — | — | |
| Dong et al. (2019)2021.04 | 38.45 | 19.45 | 35.75 | — | — | |
| TransformerDecoding strategy=beam-search (beam=4), Model architecture=Transformer base2019.05 | 37.87 | 18.92 | 35.13 | 149 | 10.1 | |
| Levenshtein TransformerTraining teacher=distillation (autoregressive teacher model), Model architecture=Transformer base2019.05 | 37.4 | 18.33 | 34.51 | 84 | 1.73 | |
| TransformerDecoding strategy=greedy, Model architecture=Transformer base2019.05 | 37.31 | 18.1 | 34.65 | 116 | 10.1 | |
| Levenshtein TransformerTraining teacher=oracle, Model architecture=Transformer base2019.05 | 36.14 | 17.14 | 34.34 | 98 | 2.32 |