List your company
ResearchFormerly WizWand

Part-of-Speech Tagging on WSJ (test)

97.78AccuracyLing et al.
95.013695.731896.4597.1682Apr 19, 2016May 21, 2017Jun 23, 2018Jul 25, 2019Aug 26, 2020Sep 27, 2021Oct 30, 2022
Updated 8mo ago

Evaluation results

MethodLinks
Ling et al.2017.1197.78—
LM-LSTM-CRFType=maximum, pre-trained word embedding=true, leverage language models=true2017.0997.59—
Adversarial POS Taggingmode=adversarial training2017.1197.58—
Ma and HovyYear=20162018.0697.55—
Yang et al.Year=20172018.0697.55—
LM-LSTM-CRFType=reported, pre-trained word embedding=true, leverage language models=true2017.0997.55—
Ma and Hovy2017.1197.55—
Yang et al.2017.1197.55—
Hashimoto et al.2017.1197.55—
BiLSTM-CRFmode=baseline2017.1197.54—
LM-LSTM-CRFType=mean, pre-trained word embedding=true, leverage language models=true2017.0997.53—
Lample et al.Year=20162018.0697.51—
Lample et al. 2016Type=maximum, pre-trained word embedding=true2017.0997.51—
Søgaard 2011Type=reported2017.0997.5—
Søgaard2017.1197.5—
SrcEmbeddings=GloVe2018.0497.5—
CLSTM+WLSTM+CRFCharacter sequence representation=LSTM, Word sequence representation=LSTM, Inference layer=CRF2018.0697.49—
CCNN+WLSTM+CRFCharacter sequence representation=CNN, Word sequence representation=LSTM, Inference layer=CRF2018.0697.46—
Ma et al. 2016Type=maximum, pre-trained word embedding=true2017.0997.46—
TriEmbeddings=GloVe2018.0497.45—
Yang et al. 2017Type=reported, pre-trained word embedding=true2017.0997.43—
Stanford*2018.0497.43—
Ma et al. 2016Type=mean, pre-trained word embedding=true2017.0997.42—
CLSTM+WCNN+CRFCharacter sequence representation=LSTM, Word sequence representation=CNN, Inference layer=CRF2018.0697.38—
MT-TriEmbeddings=GloVe2018.0497.37—
Bi-LSTMReference=Ling et al., 20152016.0497.36—
Sun 2014Type=reported2017.0997.36—
Lample et al. 2016Type=mean, pre-trained word embedding=true2017.0997.35—
CCNN+WCNN+CRFCharacter sequence representation=CNN, Word sequence representation=CNN, Inference layer=CRF2018.0697.33—
ConvnetReference=Santos and Zadrozny, 20142016.0497.32—
Collobert et al. 2011Type=reported, pre-trained word embedding=true2017.0997.29—
Collobert et al.2017.1197.29—
Manning 2011Type=reported2017.0997.28—
Manning2017.1197.28—
Toutanova et al.2017.1197.27—
bi-LSTM (w+c)Input Representation=words (w) + characters (c), Epochs=30, sigma=0.3, Pre-training=no POLYGLOT2016.0497.22—
Nochar+WLSTM+CRFCharacter sequence representation=None, Word sequence representation=LSTM, Inference layer=CRF2018.0697.2—
AsymEmbeddings=GloVe2018.0497.19—
Rei 2017Type=max, pre-trained word embedding=true, leverage language models=true2017.0997.14—
FLORS*2018.0497.11—
Nochar+WCNN+CRFCharacter sequence representation=None, Word sequence representation=CNN, Inference layer=CRF2018.0696.99—
Rei 2017Type=mean, pre-trained word embedding=true, leverage language models=true2017.0996.97—
Context-to-Vecbackbone=BERT2022.1096.9192.8
BERT+Skip-gram2022.1096.8692.3
Convnet (reimplementation)Reference=Ling et al., 20152016.0496.8—
BERTavgpooling=average2022.1096.890.76
SynGCN2022.1096.7192.23
TnT*2018.0496.57—
GloVe2022.1096.5291.84
Bi-RNNReference=Ling et al., 20152016.0495.93—
Skip-gram2022.1095.1289.03