Interested in Contributing?

Scored Systems

System Submitter System Notes Constraint Run Notes BLEU BLEU-cased TER BEER 2.0 CharactTER
Tencent ensemble system  (Details) Mr Translator
Tencent
Rerank ensemble outputs with 48 features (including t2t R2l, t2t L2R, rnn L2R, rnn R2L etc.) Back translation. Joint train with English to Chinese systems. Fine-tuning with selected data. Knowledge distillation. yes

30.8

29.3

failed

0.594

0.574

NiuTrans  (Details) NiuTrans
Northeastern University
Ensemble of 15 Transformer models with re-ranking yes Ensemble of 15 Transformer models with re-ranking

29.9

28.7

failed

0.589

0.589

Uni-NMT-Transformer-ZhEn  (Details) Unisound
Unisound AI Labs
BackTranslation + Ensemble + Rerank(ZhEn-L2R + ZhEn-R2L + EnZh-L2R + EnZh-R2L +LM) yes BackTranslation + Ensemble + Rerank(ZhEn-L2R + ZhEn-R2L + EnZh-L2R + EnZh-R2L +LM, Average rescore weight)

30.1

28.4

failed

0.590

0.589

TencentFmRD-zhen-new  (Details) Bojie Hu
TencentFmRD
Transformer, ensemble, reranking, finetune, back-translation yes Transformer, ensemble, reranking, finetune, back-translation

30.2

28.3

failed

0.593

0.576

TencentFmRD-zhen  (Details) Bojie Hu
TencentFmRD
Transformer, ensemble, reranking, finetune, back-translation yes Transformer, ensemble, reranking, finetune, back-translation

30.1

28.2

failed

0.593

0.577

NMT-SMT Hybrid  (Details) fstahlberg
University of Cambridge
MBR-based combination of neural models and SMT yes

29.0

27.7

failed

0.587

0.589

Uni-NMT Transformer  (Details) Unisound
Unisound AI Labs
BackTranslation + Ensemble + Rerank + SMT yes average weight rerank

29.3

27.7

failed

0.589

0.593

Tencent single system  (Details) Mr Translator
Tencent
Single LSTM systems with 6 layers encoders and 3 layers decoders. Back translation (ensemble) with parallel target parts and 20 million mono lingual corpus. Trained with R2l regulazition and target to source joint training. Fine-tuning with selected data from CNN and RNN classification model. yes

29.8

27.5

failed

0.585

0.583

Li Muze  (Details) Li Muze
CCNI
Ensembles of 4 averaged Transformer models with 1 zh-en R2L and 1 en-zh T2S averaged Transformer model, all the models are same as Transformer big-model, trained on the official training data with 4.5M back-translation data on the news2016&2017 data. And English vocabulary size is 3.6w BPE subwords. yes

28.5

27.4

failed

0.585

0.603

NICT  (Details) rui.wang
NICT
The same team with benjamin.marie of NICT. yes Transformer, back-translation, ensemble and reranking

28.0

26.7

failed

0.578

0.610

test_normal  (Details) wangwei
yes

27.5

26.4

failed

0.578

0.612

zzy_zh2en2  (Details) wangwei
test yes

27.3

26.1

failed

0.577

0.613

RWTH Transformer  (Details) pbahar
RWTH Aachen University
The ensemble of 4 checkpoints, back-translated data yes

27.3

26.1

failed

0.573

0.622

ForyorMT_Chinese2English  (Details) Zeng Hui
no

27.3

25.9

failed

0.580

failed

bit-zhen  (Details) bit-nmt
yes

27.2

25.8

failed

0.576

0.621

ForyorMT_Chinese2English  (Details) Zeng Hui
no

27.0

25.7

failed

0.572

failed

bit-zhen  (Details) bit-nmt
yes

26.6

25.3

failed

0.573

0.639

Wonder Woman  (Details) fansiawang
Personal
yes

27.6

25.0

failed

0.570

0.638

PERCY-trans   (Details) PERCY-sys
ATT
single yes

25.8

24.7

failed

0.570

0.648

Wonder Woman  (Details) fansiawang
Personal
yes

27.0

24.5

failed

0.566

0.638

bit-zhen  (Details) bit-nmt
yes

25.9

24.5

failed

0.568

0.721

bit-zhen  (Details) bit-nmt
yes

25.9

24.4

failed

0.567

0.723

transformer  (Details) weijia
University of Maryland
yes ensemble of 3 transformer models, reranking with r2l, t2s

25.6

24.4

failed

0.570

failed

Wonder Woman  (Details) fansiawang
Personal
yes

26.4

24.1

failed

0.569

0.666

Wonder Woman  (Details) fansiawang
Personal
yes

26.4

24.0

failed

0.569

0.666

UEDIN  (Details) XapaJIaMnu
UEDIN
yes Best deep with layer normalization and multi-head attention. Small vocabulary (18k) + ensembles trained for about 40 hours

25.1

24.0

failed

0.562

0.680

Wonder Woman  (Details) fansiawang
Personal
yes

26.4

23.8

failed

0.568

0.667

Wonder Woman  (Details) fansiawang
Personal
yes

26.4

22.0

failed

0.558

0.675

Wonder Woman  (Details) fansiawang
Personal
yes

26.4

21.6

failed

0.556

0.677

S-MT  (Details) Hongxin Shao
Shopee
yes

22.1

21.1

failed

0.546

0.697

Wonder Woman  (Details) fansiawang
Personal
yes ensemble of 4 transformer model

failed

failed

failed

0.556

0.677

yanghaocsg  (Details) yanghaocsg
dr
yes
Wonder Woman  (Details) fansiawang
Personal
yes ensemble of 4 transformer models

failed

failed

failed

0.556

0.677

A3-180  (Details) saumitray
IIIT Hyderabad
Baseline NMT system with Global Attention no

failed

failed

failed

0.447

0.915

NiuTrans  (Details) NiuTrans
Northeastern University
Ensemble of 15 Transformer models with re-ranking yes Ensemble of 15 Transformer models with re-ranking

failed

failed

failed

0.589

0.589

A3-180  (Details) saumitray
IIIT Hyderabad
Baseline NMT system with Global Attention no