Interested in Contributing?

Scored Systems

System Submitter System Notes Constraint Run Notes BLEU BLEU-cased TER BEER 2.0 CharactTER
uedin-nmt-ensemble  (Details) rsennrich
University of Edinburgh
BPE neural MT system with monolingual training data (back-translated). ensemble of 4, reranked with right-to-left model. yes

34.8

34.2

0.543

0.626

0.468

UniMelb-NMT-Transformer-BT  (Details) vhoang
The University of Melbourne, Australia
NMT with Transformer architecture (medium size, 4 heads + 4 encoder/decoder layers) enhanced with WMT'16 back-translation data. (decoding with single best system) yes

34.7

34.0

0.533

0.625

0.475

KIT-cG-mA-single 2016  (Details) thanhleha
KIT
yes

33.4

32.8

0.551

0.620

0.480

metamind-ensemble  (Details) jekbradbury
Salesforce MetaMind
Neural MT system based on Luong 2015 and Sennrich 2015, using Morfessor for subword splitting, with back-translated monolingual augmentation. Ensemble of 3 checkpoints from one run plus 1 Y-LSTM (see entry). yes

32.8

32.3

0.546

0.617

0.490

uedin-nmt-single  (Details) rsennrich
University of Edinburgh
BPE neural MT system with monolingual training data (back-translated). single model. (contrastive) yes

32.2

31.6

0.570

0.611

0.492

metamind-single  (Details) jekbradbury
Salesforce MetaMind
Neural MT system based on Luong 2015 and Sennrich 2015, using Morfessor for subword splitting, with back-translated monolingual augmentation. Single model. yes

32.1

31.6

0.553

0.616

0.495

KIT correct BT single 2016  (Details) thanhleha
KIT
yes

32.1

31.5

0.556

0.613

0.514

NYU-UMontreal-NMT-BPE-Char  (Details) jychung08
University of Montreal
BPE to Character neural machine translation system http://arxiv.org/abs/1603.06147 yes

31.3

30.8

0.583

0.600

0.511

uedin-syntax  (Details) rsennrich
University of Edinburgh
string-to-tree syntax-based SMT system, mostly corresponding to system described in http://www.aclweb.org/anthology/D/D15/D15-1248.pdf , with added News 2015 LM. yes

31.4

30.6

0.585

0.610

0.505

nmt-editdistance-hifst  (Details) fs439
University of Cambridge
Loose coupling of NMT and Hiero via edit distance transducer yes update LM to news2015, fix surnames

31.3

30.6

0.580

0.599

0.531

NYU-UMontreal-NMT-BPE-Char  (Details) jychung08
University of Montreal
BPE to Character neural machine translation system http://arxiv.org/abs/1603.06147 yes

31.1

30.6

0.585

0.599

0.511

nmt-editdistance-hifst  (Details) fs439
University of Cambridge
Loose coupling of NMT and Hiero via edit distance transducer yes

31.1

30.5

0.578

0.597

0.536

nmt-cgru-ensemble  (Details) ozancaglayan
LIUM - Le Mans University
Ensemble of 3 small models, CGRU-based NMT, subword. yes

30.7

30.1

0.605

0.599

0.516

metamind-ylstm  (Details) jekbradbury
Salesforce MetaMind
Novel neural MT system with source and target LSTM-LMs coupled to each other at the middle layer. Uses Morfessor for subword splitting, with back-translated monolingual augmentation. yes

29.8

29.3

0.575

0.598

0.511

KIT/LIMSI  (Details) niehues
KIT
Phrase-based MT system with NMT and SOUL in rescoring yes

29.7

29.1

0.594

0.600

0.533

KIT  (Details) niehues
KIT
Phrase-based MT with NMT in rescoring yes

29.7

29.0

0.598

0.602

0.533

uedin-pbt-wmt16-en-de  (Details) Matthias Huck
University of Edinburgh
Phrase-based Moses yes

29.1

28.4

0.600

0.594

0.556

Moses Phrase-Based  (Details) jhu-smt
Johns Hopkins University
Phrase-based model, word clusters for all model components (LM, OSM, LR, sparse features), neural network joint model, large cc LM yes [26-7]

29.0

28.3

0.629

0.595

0.537

uedin-pbt-wmt16-en-de-contrastive  (Details) Matthias Huck
University of Edinburgh
Phrase-based Moses (contrastive, 2015 system) yes

29.0

28.3

0.618

0.594

0.540

jhu-syntax  (Details) sding
Johns Hopkins University
yes

27.3

26.6

0.637

0.593

0.533

ParFDA  (Details) bicici
en-de ParFDA Moses phrase-based SMT system yes en-de (after the deadline)

24.5

23.9

0.680

0.569

0.590

PROMT Rule-based  (Details) Alex Molchanov
PROMT LLC
no

23.8

23.4

0.651

0.570

0.538