Constituent parsing on PTB (test)
96.48F1Yang and Tu
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Yang and TuBackbone=XLNet, year=20232024.06 | 96.48 | — | — | — | |
| bit-level parserBackbone=XLNet, Bit-width=14 bits2024.06 | 96.43 | — | — | — | |
| ZEN/XLNET + POS + CATSAEncoder=ZEN/XLNET, Predicted POS tags=true, Span Attention=Categorical2020.10 | 96.4 | 96.19 | 96.61 | — | |
| Tian et al.Backbone=XLNet2024.06 | 96.4 | — | — | — | |
| bit-level parserBackbone=XLNet, Bit-width=16 bits2024.06 | 96.4 | — | — | — | |
| Tian et al. + PoS + XLNetlargeBackbone=XLNetlarge, Auxiliary features=+PoS2026.05 | 96.4 | — | — | — | |
| MRINI ET AL. (BERT/XLNET + POS)Encoder=BERT/XLNET, POS tags=Predicted, year=20192020.10 | 96.38 | 96.53 | 96.24 | — | |
| Mrini et al.#Params=459M, Pre-trained embedding=XLNet, Extra supervision from dependency parsing data=true, Type=chart-based2020.10 | 96.38 | 96.24 | 96.53 | 58.65 | |
| LAL-Parser2019.11 | 96.38 | 96.24 | 96.53 | — | |
| Mrini et al.Backbone=XLNet2024.06 | 96.38 | — | — | — | |
| Mrini et al. + dep + PoS + XLNetlargeBackbone=XLNetlarge, Auxiliary features=+dep + PoS2026.05 | 96.38 | — | — | — | |
| bit-level parserBackbone=XLNet, Bit-width=10 bits2024.06 | 96.37 | — | — | — | |
| ZEN/XLNET + CATSAEncoder=ZEN/XLNET, Span Attention=Categorical2020.10 | 96.36 | 96.07 | 96.64 | — | |
| bit-level parserBackbone=XLNet, Bit-width=12 bits2024.06 | 96.36 | — | — | — | |
| Tian et al. + XLNetlargeBackbone=XLNetlarge2026.05 | 96.36 | — | — | — | |
| attach-juxtapose (XLNet)#Params=391M, Pre-trained embedding=XLNet, Beam search=102020.10 | 96.34 | 96.13 | 96.55 | 59.17 | |
| Yang and DengBackbone=XLNet2024.06 | 96.34 | — | — | — | |
| bit-level parserBackbone=XLNet, Bit-width=8 bits2024.06 | 96.34 | — | — | — | |
| Yang and Deng + XLNetlargeBackbone=XLNetlarge2026.05 | 96.34 | — | — | — | |
| Joint Span HPSG + XLNetDecoding=Joint, Pre-training=XLNet2019.07 | 96.33 | 96.21 | 96.46 | — | |
| ZHOU AND ZHAO (XLNET)Encoder=XLNET, year=20192020.10 | 96.33 | 96.46 | 96.21 | — | |
| ZEN/XLNET + POS + SAEncoder=ZEN/XLNET, Predicted POS tags=true, Span Attention=Normal2020.10 | 96.33 | 96.1 | 96.56 | — | |
| Zhou and Zhao#Params=374M, Pre-trained embedding=XLNet, Extra supervision from dependency parsing data=true, Type=chart-based2020.10 | 96.33 | 96.21 | 96.46 | 58.73 | |
| Zhou and Zhao (2019)Backbone=XLNet2019.11 | 96.33 | 96.21 | 96.46 | — | |
| Zhou and ZhaoBackbone=XLNet2024.06 | 96.33 | — | — | — | |
| Zhou and Zhao + dep + XLNetlargeBackbone=XLNetlarge, Auxiliary features=+dep2026.05 | 96.33 | — | — | — | |
| ZEN/XLNET + SAEncoder=ZEN/XLNET, Span Attention=Normal2020.10 | 96.31 | 96.03 | 96.58 | — | |
| BARTlarge + lexicalized in-orderBackbone=BARTlarge, Lexicalization=true, Linearization strategy=in-order2026.05 | 96.24 | — | — | — | |
| BARTlarge + lexicalized top-downBackbone=BARTlarge, Lexicalization=true, Linearization strategy=top-down2026.05 | 96.23 | — | — | — | |
| ZEN/XLNET + POSEncoder=ZEN/XLNET, Predicted POS tags=true2020.10 | 96.14 | 95.86 | 96.42 | — | |
| ZEN/XLNETEncoder=ZEN/XLNET2020.10 | 96.11 | 95.7 | 96.52 | — | |
| Yang and TuBackbone=BERT, year=20232024.06 | 96.04 | — | — | — | |
| bit-level parserBackbone=BERT, Bit-width=10 bits2024.06 | 96.03 | — | — | — | |
| bit-level parserBackbone=BERT, Bit-width=14 bits2024.06 | 96.02 | — | — | — | |
| Bottom-up Pointer NetworksEncoder=BERT, Method Type=seq2seq-based2021.10 | 96.01 | 95.83 | 96.19 | — | |
| Yang and TuBackbone=BERT, year=20222024.06 | 96.01 | — | — | — | |
| bit-level parserBackbone=BERT, Bit-width=12 bits2024.06 | 96 | — | — | — | |
| bit-level parserBackbone=BERT, Bit-width=16 bits2024.06 | 95.98 | — | — | — | |
| bit-level parserBackbone=BERT, Bit-width=8 bits2024.06 | 95.95 | — | — | — | |
| Xin et al. (2021)Encoder=BERT, Method Type=span-based2021.10 | 95.92 | 95.55 | 96.29 | — | |
| Cui et al. (2021)Encoder=BERT, Method Type=span-based2021.10 | 95.92 | 96.14 | 95.7 | — | |
| NFCPre-trained model=BERT, Evaluation setting=In-domain2021.09 | 95.92 | — | — | — | |
| Xin et al.Backbone=BERT2024.06 | 95.92 | — | — | — | |
| Cui et al.Backbone=BERT2024.06 | 95.92 | — | — | — | |
| BERT + POS + CATSAEncoder=BERT, Predicted POS tags=true, Span Attention=Categorical2020.10 | 95.86 | 95.62 | 96.09 | — | |
| Tian et al. (2020)Encoder=BERT, Method Type=span-based2021.10 | 95.86 | 95.62 | 96.09 | — | |
| Tian et al.Backbone=BERT2024.06 | 95.86 | — | — | — | |
| Tian et al. + PoS + BERTLargeBackbone=BERTLarge, Auxiliary features=+PoS2026.05 | 95.86 | — | — | — | |
| BERT + CATSAEncoder=BERT, Span Attention=Categorical2020.10 | 95.85 | 95.58 | 96.11 | — | |
| Joint Span HPSG + BERTDecoding=Joint, Pre-training=BERT2019.07 | 95.84 | 95.7 | 95.98 | — | |
| ZHOU AND ZHAO (BERT)Encoder=BERT, year=20192020.10 | 95.84 | 95.98 | 95.7 | — | |
| Zhou and Zhao#Params=349M, Pre-trained embedding=BERT, Extra supervision from dependency parsing data=true, Type=chart-based2020.10 | 95.84 | 95.7 | 95.98 | 55.84 | |
| Zhou and Zhao (2019)Backbone=BERT2019.11 | 95.84 | 95.7 | 95.98 | — | |
| Zhou and Zhao (2019)Encoder=BERT, Method Type=span-based2021.10 | 95.84 | 95.98 | 95.7 | — | |
| Zhou and ZhaoBackbone=BERT2024.06 | 95.84 | — | — | — | |
| Zhou and Zhao + dep + BERTLargeBackbone=BERTLarge, Auxiliary features=+dep2026.05 | 95.84 | — | — | — | |
| F&R [15] + RoBERTaLarge + in-orderBackbone=RoBERTaLarge, Linearization strategy=in-order2026.05 | 95.84 | — | — | — | |
| Wei et al. (2020)Encoder=BERT, Method Type=span-based2021.10 | 95.8 | 96.1 | 95.5 | — | |
| attach-juxtapose (BERT)#Params=377M, Pre-trained embedding=BERT, Beam search=102020.10 | 95.79 | 95.55 | 96.04 | 57.29 | |
| Yang and Deng (2020)Encoder=BERT, Method Type=transition-based2021.10 | 95.79 | 95.55 | 96.04 | — | |
| Yang and DengBackbone=BERT2024.06 | 95.79 | — | — | — | |
| Yang and Deng + BERTLargeBackbone=BERTLarge2026.05 | 95.79 | — | — | — | |
| F&R [15] + RoBERTaLarge + top-downBackbone=RoBERTaLarge, Linearization strategy=top-down2026.05 | 95.78 | — | — | — | |
| Kitaev and KleinPre-training=BERT (Ensemble/Large), Year=2018b2019.07 | 95.77 | 95.51 | 96.03 | — | |
| Kitaev et al. (2018)2019.11 | 95.77 | 95.51 | 96.03 | — | |
| BERT + POS + SAEncoder=BERT, Predicted POS tags=true, Span Attention=Normal2020.10 | 95.73 | 95.51 | 95.96 | — | |
| Division Span HPSG + BERTDecoding=Division, Pre-training=BERT2019.07 | 95.72 | 95.51 | 95.93 | — | |
| BERT + SAEncoder=BERT, Span Attention=Normal2020.10 | 95.72 | 95.4 | 96.04 | — | |
| Kitaev and Klein (2018)Pre-trained model=BERT, Evaluation setting=In-domain2021.09 | 95.72 | — | — | — | |
| FRIED ET AL. (BERT)Encoder=BERT, year=20192020.10 | 95.71 | — | — | — | |
| Liu and ZhangType=transition-based, Pre-trained embedding=BERT, Re-implementation source=Fried et al. [13], Beam search=102020.10 | 95.71 | — | — | 57.05 | |
| Nguyen et al. (2021a)Encoder=BERT, Method Type=seq2seq-based2021.10 | 95.7 | — | — | — | |
| Nguyen et al.Backbone=BERT2024.06 | 95.7 | — | — | — | |
| bit-level parserBackbone=XLNet, Bit-width=6 bits2024.06 | 95.7 | — | — | — | |
| Neural CRF ParserEmbeddings=BERT2020.08 | 95.69 | 95.53 | 95.85 | — | |
| Zhang et al. (2020)Encoder=BERT, Method Type=span-based2021.10 | 95.69 | 95.53 | 95.85 | — | |
| Zhang et al.Backbone=BERT2024.06 | 95.69 | — | — | — | |
| BARTbase + lexicalized top-downBackbone=BARTbase, Lexicalization=true, Linearization strategy=top-down2026.05 | 95.69 | — | — | — | |
| Liu and Zhang (2017)Pre-trained model=BERT, Evaluation setting=In-domain2021.09 | 95.65 | — | — | — | |
| BARTbase + lexicalized in-orderBackbone=BARTbase, Lexicalization=true, Linearization strategy=in-order2026.05 | 95.61 | — | — | — | |
| BERT + POSEncoder=BERT, Predicted POS tags=true2020.10 | 95.6 | 95.27 | 95.92 | — | |
| Kitaev and Klein + BERTPre-training=BERT, Year=2018b2019.07 | 95.59 | 95.46 | 95.73 | — | |
| Kitaev et al. [2019]Embeddings=BERT2020.08 | 95.59 | 95.46 | 95.73 | — | |
| KITAEV ET AL. (BERT)Encoder=BERT, year=20192020.10 | 95.59 | 95.46 | 95.73 | — | |
| Kitaev et al.#Params=342M, Pre-trained embedding=BERT, Type=chart-based2020.10 | 95.59 | 95.73 | 95.46 | — | |
| Kitaev et al. (2019)Encoder=BERT, Method Type=span-based2021.10 | 95.59 | 95.73 | 95.46 | — | |
| Kitaev et al.Backbone=BERT2024.06 | 95.59 | — | — | — | |
| Kitaev et al. + BERTLargeBackbone=BERTLarge2026.05 | 95.59 | — | — | — | |
| T5large + lexicalized top-downBackbone=T5large, Lexicalization=true, Linearization strategy=top-down2026.05 | 95.55 | — | — | — | |
| BERTEncoder=BERT2020.10 | 95.54 | 95.17 | 95.91 | — | |
| Nguyen et al. (2020)Encoder=BERT, Method Type=span-based2021.10 | 95.48 | — | — | — | |
| T5large + lexicalized in-orderBackbone=T5large, Lexicalization=true, Linearization strategy=in-order2026.05 | 95.45 | — | — | — | |
| BARTlarge + lexicalized bottom-upBackbone=BARTlarge, Lexicalization=true, Linearization strategy=bottom-up2026.05 | 95.36 | — | — | — | |
| Joint Span HPSG + ELMoDecoding=Joint, Pre-training=ELMo2019.07 | 95.22 | 95.04 | 95.39 | — | |
| BNPcitation=KITAEV AND KLEIN, 20182020.10 | 95.17 | 94.89 | 95.46 | — | |
| Kitaev and Klein + ELMoPre-training=ELMo, Year=2018a2019.07 | 95.13 | 94.85 | 95.4 | — | |
| Kitaev and Klein [2018]Embeddings=ELMo2020.08 | 95.13 | 94.85 | 95.4 | — | |
| KITAEV AND KLEINyear=20182020.10 | 95.13 | 94.85 | 95.4 | — | |
| Kitaev and Klein#Params=107M, Pre-trained embedding=ELMo, Type=chart-based2020.10 | 95.13 | 94.85 | 95.4 | 53.06 | |
| Kitaev and Klein (2018)2019.11 | 95.13 | 94.85 | 95.4 | — |