List your company
ResearchFormerly WizWand

Machine Translation (Chinese-to-English) on NIST 2003 (MT-03)

50.3BLEUManifold
31.715236.540141.36546.1899Jun 10, 2018Mar 8, 2019Dec 4, 2019Aug 31, 2020May 29, 2021Feb 24, 2022Nov 23, 2022
Updated 8mo ago

Evaluation results

MethodLinks
ManifoldTotal Parameters=83M, Embedding Parameters=46M2022.1150.3
AdvAugLoss Config.=L_aut + L_adv, Extra corpora=true2020.0649.81
StrokeNetTotal Parameters=59M, Embedding Parameters=15M2022.1149.8
AdvAugLoss Config.=L_clean + L_adv, Extra corpora=false2020.0648.73
AdvAugTotal Parameters=95M, Embedding Parameters=39M2022.1148
AdvAugLoss Config.=L_aut + L_adv, Extra corpora=false2020.0647.96
VanillaTotal Parameters=80M, Embedding Parameters=36M2022.1147.5
Sennrich et al.Loss Config.=not elaborated, Extra corpora=true2020.0647.1
Edunov et al.Loss Config.=not elaborated, Extra corpora=true2020.0646.93
AdvGenTotal Parameters=95M, Embedding Parameters=39M2022.1146.5
Cheng et al.Loss Config.=not elaborated, Extra corpora=false2020.0646.48
AdvAugLoss Config.=L_aut, Extra corpora=false2020.0646.13
Sato et al.Loss Config.=not elaborated, Extra corpora=false2020.0645.02
AdvAugLoss Config.=L_mixup, Extra corpora=false2020.0644.81
Miyato et al.Loss Config.=not elaborated, Extra corpora=false2020.0644.68
Vaswani et al.Loss Config.=L_clean, Extra corpora=false2020.0644.55
Shared-PrivateTotal Parameters=63M, Embedding Parameters=19M2022.1142
SAWRBackbone=Transformer, Heads=8, Layers=6, Hidden dimension=5122019.0541.63
Tree-RNNBackbone=Transformer, Heads=8, Layers=6, Hidden dimension=5122019.0541.24
Tree-LinearizationBackbone=Transformer, Heads=8, Layers=6, Hidden dimension=5122019.0541.12
TransformerBackbone=Transformer, Heads=8, Layers=6, Hidden dimension=5122019.0540.45
RNNSEARCH-COVERAGE + D_RNN + O_CDR# Para.=+1.20M, Speed=0.7K2018.1138.62
SAWRsyntax-integrated=true2019.0538.42
Seq2Seq + SACTMechanism=Self-Adaptive Control of Temperature2018.0838.16
Tree-RNNsyntax-integrated=true2019.0538.12
Seq2Seq+Attention+DeconvDec2018.0638.04
RNNSEARCH-COVERAGE# Para.=+1.03M, Speed=1.2K2018.1138.04
Tree-Linearizationsyntax-integrated=true2019.0537.95
RNNSEARCH + O_CHRF3# Para.=+0M, Speed=1.1K2018.1137.91
RNNSEARCH + D_RNN + O_CDR# Para.=+0.17M, Speed=0.8K2018.1137.61
RNNSEARCH + O_CDR# Para.=+0M, Speed=1.0K2018.1137.42
RNNSEARCH + MRT_BLEU# Para.=+0M, Speed=0.3K2018.1137.32
RNNSEARCH + D_CNN# Para.=+0.23M, Speed=1.0K2018.1137.11
RNNSEARCH + O_BLEU# Para.=+0M, Speed=1.1K2018.1137.02
RNNSEARCH + D_RNN# Para.=+0.17M, Speed=1.2K2018.1136.96
Baseline2019.0536.44
MemDec2018.0636.16
MemDecType=Attention-based Seq2Seq with external memory2018.0836.16
RNNSEARCH# Para.=86.7M, Speed=1.4K2018.1136
Chen et al. (2017b)2019.0535.91
Chen et al. (2017a)2019.0535.64
Seq2Seq+Attention2018.0635.32
Seq2Seq2018.0835.32
InterAtten2018.0635.09
Li et al. (2017)2019.0534.9
Coverage2018.0634.49
CoverageType=Attention-based Seq2Seq with coverage model2018.0834.49
Lattice2018.0634.32
RNNSearch2018.0633.08
RNNSearchType=Attention-based Seq2Seq2018.0833.08
Moses2018.0632.43
MosesType=Phrase-based translation, Language Model=4-gram2018.0832.43