RST Discourse Parsing on RST-DT Parseval (test)
88.3Span (S) ScoreHuman
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Human2021.02 | 88.3 | 77.3 | 65.4 | 64.7 | — | — | — | — | |
| Kobayashi et al.sentence boundary features=true, paragraph boundary features=true2021.02 | 87 | 74.6 | 60 | — | — | — | — | — | |
| LSTM (dynamic)paragraph boundary features=true, oracle=dynamic2021.02 | 86.6 | 73.7 | 61.5 | 60.9 | — | — | — | — | |
| LSTM (static)paragraph boundary features=true, oracle=static2021.02 | 86.4 | 73.4 | 60.8 | 60.3 | — | — | — | — | |
| Yu et al. (static)paragraph boundary features=true, oracle=static2021.02 | 85.8 | 72.6 | 59.5 | 59 | — | — | — | — | |
| Yu et al. (dynamic)oracle=dynamic2021.02 | 85.6 | 72.9 | 59.8 | 59.3 | — | — | — | — | |
| Yu et al. (1 run)paragraph boundary features=true2021.02 | 85.5 | 73.1 | 60.2 | 59.9 | — | — | — | — | |
| Transformer (dynamic)paragraph boundary features=true, oracle=dynamic2021.02 | 85.5 | 72.3 | 60.5 | 59.9 | — | — | — | — | |
| Transformer (static)paragraph boundary features=true, oracle=static2021.02 | 85.2 | 72 | 60.3 | 59.6 | — | — | — | — | |
| Feng and Hirstsentence boundary features=true2021.02 | 84.3 | 69.4 | 56.9 | 56.2 | — | — | — | — | |
| LSTM (dynamic) w/o boundaryparagraph boundary features=false, oracle=dynamic2021.02 | 83.6 | 70.4 | 58.8 | 58.2 | — | — | — | — | |
| LSTM (static) w/o boundaryparagraph boundary features=false, oracle=static2021.02 | 83.2 | 70.4 | 58.4 | 57.9 | — | — | — | — | |
| Surdeanu et al.sentence boundary features=true2021.02 | 82.6 | 67.1 | 55.4 | 54.9 | — | — | — | — | |
| Joty et al.2021.02 | 82.6 | 68.3 | 55.8 | 54.4 | — | — | — | — | |
| Hayashi et al.2021.02 | 82.6 | 66.6 | 54.6 | 54.3 | — | — | — | — | |
| Li et al.2021.02 | 82.2 | 66.5 | 51.4 | 50.6 | — | — | — | — | |
| Ji and Eisensteinsentence boundary features=true2021.02 | 82 | 68.2 | 57.8 | 57.6 | — | — | — | — | |
| Braud et al.2021.02 | 81.3 | 68.1 | 56.3 | 56 | — | — | — | — | |
| Human AgreementGold segmentation=true2021.05 | 78.7 | 66.8 | 57.1 | 55 | 88.7 | 77.3 | 65.4 | — | |
| RST-Parser (with XLNet)Embeddings=XLNet, Pretrained models=true, Gold segmentation=true2021.05 | 74.3 | 64.3 | 51.6 | 50.2 | 87.6 | 76 | 61.8 | — | |
| Yu et al. (2018)Handcrafted features=true, Pretrained models=true, Gold segmentation=true2021.05 | 71.4 | 60.3 | 49.2 | 48.1 | 85.5 | 73.1 | 60.2 | — | |
| RST-Parser (with GloVe)Embeddings=GloVe, Gold segmentation=true2021.05 | 71.1 | 59.6 | 47.7 | 46.8 | — | — | — | — | |
| Feng and Hirst (2014)Handcrafted features=true, Gold segmentation=true2021.05 | 68.6 | 55.9 | 45.8 | 44.6 | — | — | — | — | |
| RST-ParserPre-trained model=XLNet2021.05 | 68.4 | 59.1 | 47.8 | 46.6 | — | — | — | — | |
| Zhang et al. (2020)Handcrafted features=true, Gold segmentation=true2021.05 | 67.2 | 55.5 | 45.3 | 44.3 | — | — | — | — | |
| Joty et al. (2015)Handcrafted features=true, Gold segmentation=true2021.05 | 65.1 | 55.5 | 45.1 | 44.3 | — | — | — | — | |
| Li et al. (2016)Handcrafted features=true, Gold segmentation=true2021.05 | 64.5 | 54 | 38.1 | 36.6 | — | — | — | — | |
| Ji and Eisenstein (2014)Handcrafted features=true, Gold segmentation=true2021.05 | 64.1 | 54.2 | 46.8 | 46.3 | — | — | — | — | |
| RST-ParserPre-trained model=GloVe2021.05 | 63.8 | 53 | 43.1 | 42.1 | — | — | — | — | |
| Braud et al. (2017)External cross-lingual features=true, Gold segmentation=true2021.05 | 62.7 | 54.5 | 45.5 | 45.1 | — | — | — | — | |
| Zhang et al.2021.05 | 62.3 | 50.1 | 40.7 | 39.6 | — | — | — | — | |
| Braud et al. (2016)Gold segmentation=true2021.05 | 59.5 | 47.2 | 34.7 | 34.3 | — | — | — | — | |
| Kobayashi et al. (2020)Handcrafted features=true, Pretrained models=true, Gold segmentation=true2021.05 | — | — | — | — | 87 | 74.6 | 60 | — | |
| Wang et al. (2017)Handcrafted features=true, Pretrained models=true, Gold segmentation=true2021.05 | — | — | — | — | 86 | 72.4 | 59.7 | — |