Interested in Contributing?

Scored Systems

System Submitter System Notes Constraint Run Notes BLEU BLEU-cased TER BEER 2.0 CharactTER
SogouKnowing-nmt  (Details) Sogou Knowing
Sogou
Ensemble deep NMT models, rerank with NMT right-2-left, t2s model, NGramLM, BPE, etc yes Ensemble deep NMT models, rerank with NMT right-2-left, t2s model, NGramLM, BPE, etc

27.2

26.4

0.662

0.570

0.599

xmunmt_ensemble_zh-en  (Details) Zhixing Tan
Xiamen University
Ensemble 4 models, reranked with 1 r2l model, BPE, CWMT+UN+NC, news2016 synthetic yes

27.0

26.0

0.645

0.566

0.605

NRC-2017-zh2en  (Details) NRC-CNRC
National Research Council Canada
yes BPE, back-translate 20M, finetune with selected data, ensemble 14 NMT models, rerank with IBM models, big LMs, and NNJMs.

26.9

25.8

0.662

0.562

0.632

uedin-nmt  (Details) barry
University of Edinburgh
Nematus deep model, layer norm, bpe news2016 synthetic. yes Ensemble for 3 l-r, reranked with ensemble of 3 r-l.

26.6

25.7

0.657

0.565

0.616

bit-zhen  (Details) bit-nmt
yes

26.7

25.6

0.631

0.568

0.620

SogouKnowing-nmt  (Details) Sogou Knowing
Sogou
Single deep NMT model, reranked with NMT variants and other featues, BPE, etc. yes Single deep NMT model, reranked with NMT variants and other featues, BPE, etc.

25.2

24.0

0.679

0.557

0.680

xmunmt_single_zh-en  (Details) Zhixing Tan
Xiamen University
single model, BPE, CWMT+UN+NC (contrastive) yes

24.3

23.4

0.670

0.552

0.634

rabbit  (Details) rabbit
rabbit
yes

24.2

23.1

0.663

0.554

0.650

uedin-nmt  (Details) barry
University of Edinburgh
Nematus deep model, layer norm, bpe, news2016 synthetic. yes Single best

23.8

22.9

0.680

0.551

0.636

zhang-zh-en-test  (Details) jingjingz
no

24.4

22.6

0.773

0.530

0.671

zhang-zh-en-test  (Details) jingjingz
no 4-base-trans

24.3

22.6

0.767

0.532

0.663

zhang-zh-en-test  (Details) jingjingz
no big

24.1

22.4

0.780

0.529

0.670

CASICT-DCU_NMT  (Details) peerachet
Institute Of Computing Technology, Chinese Academy Of Sciences
Neural Machine Translation (RNNSearch) Training Data : CWMT+WMT (In cooperation with School of Computing, Dublin City University) yes NMT, RNNSearch Training Data : CWMT+UN

23.4

22.3

0.676

0.538

0.733

test_zh2en  (Details) serenade
Peking University
yes

24.0

21.8

0.673

0.545

0.642

zhang-zh-en-test  (Details) jingjingz
no transformer-70050-base/eval-5-emsemble

23.5

21.8

0.768

0.529

0.675

zhang-zh-en-test  (Details) jingjingz
no avg

23.4

21.7

0.770

0.529

0.674

zhang-zh-en-test  (Details) jingjingz
no 59000

23.4

21.7

0.762

0.528

0.693

test_zh2en  (Details) serenade
Peking University
yes

23.1

21.7

0.666

0.543

0.675

test_zh2en  (Details) serenade
Peking University
yes

23.1

21.7

0.666

0.543

0.675

zhang-zh-en-test  (Details) jingjingz
no merge delete &

23.4

21.7

0.764

0.531

0.674

zhang-zh-en-test  (Details) jingjingz
no 63000

23.3

21.6

0.761

0.528

0.691

zhang-zh-en-test  (Details) jingjingz
no 58001

23.3

21.6

0.763

0.528

0.693

zhang-zh-en-test  (Details) jingjingz
no 61000

23.2

21.5

0.761

0.527

0.696

zhang-zh-en-test  (Details) jingjingz
no 70050

23.1

21.4

0.764

0.526

0.695

afrl-mitll-zhen-opennmt  (Details) jeremy.gwinnup
AFRL
8 ensemble system combination, 4 epochs from 1000 hidden, 600 embedding, brnn and 4 epochs from 500 hidden, 500 embedding, rnn yes

22.1

21.3

failed

0.539

0.660

zhang-zh-en-test  (Details) jingjingz
no ensemble

22.7

21.0

0.776

0.525

0.687

ROCMT  (Details) RocMT
University of Rochester
Single system layer normalization + etc unsupervised word and subword(BPE) segmentation minimum risk tuning rnnlm + kenlm, smt rescoring postprocessing no backtranslations yes

21.6

20.8

0.716

0.535

0.660

jhu-nmt  (Details) sding
Johns Hopkins University
Nematus, BPE, CWMT + UN, 4 ensemble yes

21.3

20.5

0.711

0.532

0.666

zhang-zh-en-test  (Details) jingjingz
no transformer-30000

21.9

20.3

0.764

0.518

0.723

zhang-zh-en-test  (Details) jingjingz
no 200000

21.2

19.7

0.777

0.516

0.721

test_zh2en  (Details) serenade
Peking University
yes

21.4

19.5

failed

0.531

0.660

NMT Model Average Multi-Cards  (Details) songkai.sk
anonymous
Baseline System yes

20.1

19.0

0.818

0.509

0.851

Oregon State University S  (Details) cosmmb
Oregon State University
two layers NMT, bpe, decoding search yes

19.4

18.5

0.722

failed

failed

Oregon State University R  (Details) cosmmb
Oregon State University
two layers NMT, bpe, decoding search yes

19.2

18.4

0.722

0.519

0.687

anonymous_test  (Details) heureux
yes SMT baseline systems

18.4

17.3

0.760

0.515

0.720

zhang-zh-en-test  (Details) jingjingz
no merge-trans-seq

18.5

17.0

0.852

0.492

0.759

PROMT SMT  (Details) Alex Molchanov
PROMT LLC
no

17.4

16.5

0.773

0.520

0.696

test_zh2en  (Details) serenade
Peking University
yes

17.2

16.1

0.710

0.488

0.869

UU-HNMT  (Details) gtang
Uppsala University
Helsinki Neural Machine Translation system yes HNMT with hybrid encoder and character-level decoder, ensemble of 5 models.

16.8

15.9

0.754

0.499

0.763

NMT Model Average Multi-Cards  (Details) songkai.sk
anonymous
Baseline System yes

20.2

15.9

0.822

0.493

0.868

ParFDA  (Details) bicici
ParFDA Moses Phrase-based SMT system yes zh-en (zh segmented with Stanford segmenter)

13.3

12.4

0.839

failed

failed

ParFDA  (Details) bicici
ParFDA Moses Phrase-based SMT system yes zh-en (zh segmented with Stanford segmenter, 2.8 million training sentences)

13.0

12.1

0.841

failed

failed

zhang-zh-en-test  (Details) jingjingz
no seq-5-merge

7.8

6.8

1.054

0.417

0.865

zhang-zh-en-test  (Details) jingjingz
no 60001.seq

7.8

6.8

1.059

0.417

0.867

att-rnn  (Details) zzz123
student
no

failed

failed

failed

0.000

0.000