Interested in Contributing?

Scored Systems

System Submitter System Notes Constraint Run Notes BLEU BLEU (11b) BLEU-cased BLEU-cased (11b) TER z BLEU-cased-norm
uedin-stanford-unconstrained  (Details) heafield
Stanford
Unconstrained. Edinburgh system with CommonCrawl on top. no

21.0

21.0

20.6

20.6

0.705

eubridge  (Details) freitag
RWTH Aachen University
system combination of the eu-bridge project yes better single systems (primary)

21.0

21.0

20.6

20.6

0.701

eubridge  (Details) freitag
RWTH Aachen University
system combination of the eu-bridge project yes alternative

21.0

21.0

20.5

20.5

0.703

uedin-syntax  (Details) rsennrich
University of Edinburgh
Moses GHKM syntax with ParZu and target-side compound splitting. yes

20.6

20.6

20.1

20.1

0.699

uedin-wmt14-en-de  (Details) Nadir
University of Edinburgh
Phrase-based Moses yes

20.5

20.5

20.1

20.1

0.708

Stanford  (Details) julia
Stanford
Phrase-based system with word class language model and features, plus a giant language model trained on German commoncrawl no Finished tuning and chose best tune iteration weights

20.9

20.9

20.0

20.0

0.710

Stanford  (Details) julia
Stanford
Phrase-based system with word class language model and features yes

20.7

20.7

19.9

19.9

0.712

KIT primary  (Details) eunah.cho
KIT
NMT, BPE, rescore using five models yes

20.0

20.0

19.5

19.5

0.708

uedin-syntax-contrastive  (Details) rsennrich
University of Edinburgh
Moses GHKM syntax with ParZu. baseline (contrastive). [not pure baseline: has compound splitting and modified label set] yes

19.9

19.9

19.5

19.5

0.718

UU-opus-ende  (Details) jorgtied
University of Helsinki
no phrase-based with preordered alignment and OPUS data

19.3

19.3

19.0

19.0

0.728

UU-ende  (Details) jorgtied
University of Helsinki
yes phrase-based with preordered alignment

19.3

19.3

18.9

18.9

0.729

uedin-syntax-wmt14-en-de-contrastive  (Details) Matthias Huck
University of Edinburgh
Moses GHKM Syntax contrastive yes uedin-syntax English-to-German contrastive, extraction based on syntactic annotation from the German Berkeley Parser, but SCFG rules with a single generic non-terminal instead of syntactic labels, plus rules added from plain phrase-based extraction

19.2

19.2

18.8

18.8

0.733

UU-opus-ende  (Details) jorgtied
University of Helsinki
no baseline with OPUS (filtered PT)

19.1

19.1

18.7

18.7

0.728

UU-Docent  (Details) sara.stymne
Uppsala University
yes Docent-based system with POS-based phrase reordering system. Primary system

19.4

19.4

18.7

18.7

0.723

uedin-syntax-wmt14-en-de-contrastive  (Details) Matthias Huck
University of Edinburgh
Moses GHKM Syntax contrastive yes uedin-syntax English-to-German contrastive, based on syntactic annotation from BitPar

19.0

19.0

18.6

18.6

0.734

uedin-syntax-wmt14-en-de-contrastive  (Details) Matthias Huck
University of Edinburgh
Moses GHKM Syntax contrastive yes uedin-syntax English-to-German contrastive, based on syntactic annotation from the German Berkeley Parser

19.0

19.0

18.6

18.6

0.732

UU-Docent  (Details) sara.stymne
Uppsala University
yes Contrastive run with Moses

19.3

19.3

18.6

18.6

0.725

UU-opus-ende  (Details) jorgtied
University of Helsinki
no preordered phrase-based with postordering, WMT + OPUS

18.9

18.9

18.5

18.5

0.729

UU-ende  (Details) jorgtied
University of Helsinki
yes simple baseline (filtered PT)

18.7

18.7

18.4

18.4

0.737

UU-ende  (Details) jorgtied
University of Helsinki
yes phrase-based, preordered with post ordering step

18.6

18.6

18.3

18.3

0.735

uedin-syntax-wmt14-en-de-contrastive  (Details) Matthias Huck
University of Edinburgh
Moses GHKM Syntax contrastive yes uedin-syntax English-to-German contrastive, based on syntactic annotation from the German Stanford Parser

18.7

18.7

18.3

18.3

0.737

uedin-syntax-wmt14-en-de-contrastive  (Details) Matthias Huck
University of Edinburgh
Moses GHKM Syntax contrastive yes uedin-syntax English-to-German contrastive, extraction based on syntactic annotation from the German Berkeley Parser, but SCFG rules with a single generic non-terminal instead of syntactic labels

18.6

18.6

18.2

18.2

0.743

CimS-CORI  (Details) Cap/Weller/Ramm
CIS, University of Munich; IMS, University of Stuttgart
Phrase-based Moses, 2-step translation process: translate into lemmatised German, with split compounds. Then, merge compounds together and generate fully inflected forms with SMOR. Enables creation of new compounds and unseen inflectional variants. Is primary. yes

18.4

18.4

17.8

17.8

0.735

CimS-RI   (Details) Cap/Weller/Ramm
CIS, University of Munich; IMS, University of Stuttgart
Phrase-based Moses, 2-step translation process: translate into lemmatised German, then generate fully inflected forms with SMOR, enables creation of unseen inflected forms yes

18.3

18.3

17.8

17.8

0.740

CimS Syntax-based Source-side reordering  (Details) Cap/Weller/Ramm
CIS, University of Munich; IMS, University of Stuttgart
String to tree syntax, GHKM Moses, source-side reordering, two-step translation process: translation into lemmatised German, then generation of fully inflected forms using SMOR yes

18.1

18.1

17.6

17.6

0.746

FDA5-DCU  (Details) bicici
Feature Decay Algorithms (FDA) 5 and Moses RELEASE-2.1 Instance selection with FDA5 and using Moses phrase based system to decode. yes en-de (after the deadline)

17.2

17.2

16.8

16.8

0.753

CimS-RIVE  (Details) Cap/Weller/Ramm
CIS, University of Munich; IMS, University of Stuttgart
Phrase-based Moses, 2-step translation process: translate into lemmatised German, then generate fully inflected forms with SMOR, enables creation of unseen inflected forms. This variant includes verbal inflection prediction. This is work in progress. yes

17.2

17.2

16.6

16.6

0.753

PROMT Hybrid  (Details) Alex Molchanov
PROMT LLC
no

16.8

16.8

16.5

16.5

0.741

TTT contrastive  (Details) Quernheim/Cap
IMS, University of Stuttgart; CIS, University of Munich
Tree-to-tree translation without unknown word handling. Fallback is standard Moses phrase-based yes

17.0

17.0

16.4

16.4

0.770

TTT - CimS-RI  (Details) Quernheim/Cap
IMS, University of Stuttgart; CIS, University of Munich
Two step translation process: Tree-to-Tree tanslation into lemmatised German (without unknown word handling), then generation of inflected forms using SMOR (as described in the CimS-RI system). Fallback is CimS-RI. yes

16.4

16.4

15.8

15.8

0.773

TTT - CimS-CORI  (Details) Quernheim/Cap
IMS, University of Stuttgart; CIS, University of Munich
Two step translation process: Tree-to-Tree translation (without unknown word handling) into lemmatised German with split compounds. Then, merge compounds together (using the target-parse tree structure) and generate fully inflected forms with SMOR (as described in the CimS-RI system). Fallback is CimS-CORI. yes

16.3

16.3

15.7

15.7

0.770

CimS-CORIVE  (Details) Cap/Weller/Ramm
CIS, University of Munich; IMS, University of Stuttgart
Phrase-based Moses, 2-step translation process: translate into lemmatised German, with split compounds. Then, merge compounds together and generate fully inflected forms with SMOR. This variant includes verbal inflection prediction. Enables creation of new compounds and unseen inflectional variants. This is work in progress. yes

16.0

16.0

15.4

15.4

failed

TTT - CimS-RI  (Details) Quernheim/Cap
IMS, University of Stuttgart; CIS, University of Munich
Two step translation process: Tree-to-Tree tanslation into lemmatised German (without unknown word handling), then generation of inflected forms using SMOR (as described in the CimS-RI system). Fallback is CimS-RI. yes For this run, the syntactic information from the target parse tree was used for re-inflection (instead of CimS-RI models). Fallback is still CimS-RI

15.8

15.8

15.3

15.3

0.780

TTT - CimS-CORI  (Details) Quernheim/Cap
IMS, University of Stuttgart; CIS, University of Munich
Two step translation process: Tree-to-Tree translation (without unknown word handling) into lemmatised German with split compounds. Then, merge compounds together (using the target-parse tree structure) and generate fully inflected forms with SMOR (as described in the CimS-RI system). Fallback is CimS-CORI. yes For this run, the syntactic information from the target parse tree was used for re-inflection (instead of CimS-RI models). Fallback is still CimS-CORI

15.7

15.7

15.1

15.1

0.776

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

14.4

14.4

14.1

14.1

0.775

Templatic  (Details) hmghaly
CUNY
no checking formatting

failed

failed

failed

failed

failed

failed

Templatic  (Details) hmghaly
CUNY
no testing file format

failed

failed

failed

failed

failed

failed