Interested in Contributing?

Scored Systems

System Submitter System Notes Constraint Run Notes BLEU BLEU-cased TER BEER 2.0 CharactTER
abumatran-combo-en-fi  (Details) atoral
Dublin City University
Combination of our 2 best individual systems (NMT and SMT) tuned on character sequences yes Reranked with left-to-right and right-to-left RNN LMs

17.8

17.4

0.726

0.533

0.622

abumatran-nmt-en-fi  (Details) vmsanchez
Prompsit Language Engineering/ABuMaTran
Neural MT system built from parallel data + backtranslated news, rule-based morphological segmentation on target side, post-processing to keep named entities yes Ensemble of 4 models

17.5

17.2

0.738

0.532

0.612

UH opus  (Details) jorgtied
University of Helsinki
added data from OPUS no non-factored including back-translated news (after the deadline)

17.1

16.4

0.752

0.533

0.629

UH opus  (Details) jorgtied
University of Helsinki
added data from OPUS no without factors

17.0

16.2

0.746

0.531

0.634

UH opus  (Details) jorgtied
University of Helsinki
added data from OPUS no factored

16.6

15.7

0.744

0.530

0.634

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

15.4

15.1

0.771

0.504

0.629

abumatran-pbsmt-en-fi  (Details) atoral
Dublin City University
Phrase-based SMT (OSM, biNLM, 3 reordering models), rule-based morphological segmentation on target side joined with BPE yes Reranked with left-to-right and right-to-left RNN LMs

14.8

14.6

0.763

0.518

0.639

abumatran-pbsmt-en-fi  (Details) vmsanchez
Prompsit Language Engineering/ABuMaTran
Phrase-based SMT (OSM, biNLM, 3 reordering models), rule-based morphological segmentation on target side joined with BPE yes

14.6

14.3

0.767

0.518

0.636

Moses Phrase-Based  (Details) jhu-smt
Johns Hopkins University
Phrase-based model, word clusters for all model components (LM, OSM, LR, sparse features), large cc LM -- no neural network joint model, no handling of morphology yes [34-5]

14.1

13.8

0.778

0.509

0.649

UH factored  (Details) jorgtied
University of Helsinki
yes including back-translated news (after the deadline)

14.3

13.6

0.765

0.507

0.672

UH factored  (Details) jorgtied
University of Helsinki
yes

13.5

12.8

0.784

0.506

0.661

UH pbsmt  (Details) jorgtied
University of Helsinki
yes

13.3

12.7

0.782

0.506

0.664

UH docent  (Details) jorgtied
University of Helsinki
yes initialisation

13.2

12.6

0.791

0.508

0.658

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

12.8

12.5

0.789

0.499

0.675

UH docent  (Details) jorgtied
University of Helsinki
yes gappy bilingual LM, intermediate result

13.0

12.4

0.795

0.508

0.664

jhu-hltcoe  (Details) post
Johns Hopkins University
Hiero model, Brown-cluster LM (k=1000), large CC LM yes 11-recased

13.1

11.9

0.793

0.500

0.653

UUT  (Details) cap
Uppsala Universitet
Phrase-based Moses, Lemmatisation, Compound Splitting, using Pseudo-Prepositions yes head inflection

12.2

11.6

0.793

0.501

0.662

UUT  (Details) cap
Uppsala Universitet
Phrase-based Moses, Lemmatisation, Compound Splitting, using Pseudo-Prepositions yes large LM

12.1

11.6

0.794

0.501

0.662

AaltoMorphsRescoredPredictor  (Details) Stig-Arne Grönroos
Aalto University
Omorfi restricted morfessor morph segmentation, Phrase-based Moses, MERT, theanolm rescoring, neural morph-boundary predictor postprocessing. yes

11.8

11.6

0.793

0.504

0.641

UUT  (Details) cap
Uppsala Universitet
Phrase-based Moses, Lemmatisation, Compound Splitting, using Pseudo-Prepositions yes medium LM

11.3

10.8

0.857

failed

failed

jhu-hltcoe  (Details) post
Johns Hopkins University
Hiero model, Brown-cluster LM (k=1000), large CC LM yes 11

13.1

10.1

0.808

0.488

0.655