Interested in Contributing?
- Check out available resources.
- Create an account and start submitting your own systems.
Scored Systems
| System | Submitter | System Notes | Run Notes | BLEU | BLEU (11b) | BLEU-cased | BLEU-cased (11b) | TER |
|---|---|---|---|---|---|---|---|---|
| uedin-wmt11-en-es (Details) | Edinburgh University of Edinburgh |
Run 26 (normalised) |
31.9 |
31.9 |
30.8 |
30.8 |
0.578 |
|
| uedin-wmt11-en-es (Details) | Edinburgh University of Edinburgh |
Run 26 |
31.9 |
31.9 |
30.8 |
30.8 |
0.578 |
|
| UCH-UPV-2011 (Details) | pakozm Universidad CEU-Cardenal Herrera and Universitat Politécnica de València |
Trained on Europarl+NewsCommentary+News+UnitedNations. Combination of independent phrase-tables over each corpus, plus Neural Network Language Models. NO LDC DATA | Linear combination of phrase-tables and rescoring with 6gram Nueral Network Language Model. PRIMARY SUBMISSION |
31.5 |
31.5 |
29.4 |
29.4 |
0.579 |
| UCH-UPV-2011 (Details) | pakozm Universidad CEU-Cardenal Herrera and Universitat Politécnica de València |
Trained on Europarl+NewsCommentary+News+UnitedNations. Combination of independent phrase-tables over each corpus, plus Neural Network Language Models. NO LDC DATA | Smoothing phrase tables by counting values (similar to resampling), plus NNLMs. CONTRASTIVE SUBMISSION |
failed |
31.4 |
failed |
29.3 |
0.580 |
| UU-baseline (Details) | jorgtied Uppsala University |
Phrase-based SMT (constrained) | standard PBSMT system, EP+News+UN (no LDC). |
31.0 |
31.0 |
29.8 |
29.8 |
0.583 |
| UCH-UPV-2011 (Details) | pakozm Universidad CEU-Cardenal Herrera and Universitat Politécnica de València |
Trained on Europarl+NewsCommentary+News+UnitedNations. Combination of independent phrase-tables over each corpus, plus Neural Network Language Models. NO LDC DATA | Smoothing phrase tables by counting CONTRASTIVE SUBMISSION |
30.9 |
30.9 |
28.9 |
28.9 |
0.585 |
| UCH-UPV-2011 (Details) | pakozm Universidad CEU-Cardenal Herrera and Universitat Politécnica de València |
Trained on Europarl+NewsCommentary+News+UnitedNations. Combination of independent phrase-tables over each corpus, plus Neural Network Language Models. NO LDC DATA | Linear phrase-tables combination. CONTRASTIVE SUBMISSION |
30.9 |
30.9 |
28.8 |
28.8 |
0.588 |
| UCH-UPV-2011 (Details) | pakozm Universidad CEU-Cardenal Herrera and Universitat Politécnica de València |
Trained on Europarl+NewsCommentary+News+UnitedNations. Combination of independent phrase-tables over each corpus, plus Neural Network Language Models. NO LDC DATA | Phrase-based baseline system. Concatenation of all available corpora. CONTRASTIVE SUBMISSION |
30.5 |
30.5 |
28.5 |
28.5 |
0.589 |
| UoW (Details) | wilkeraziz University of Wolverhampton |
Tree-to-string model trained using the entire Europarl-v6. The English side was shallow parsed and annotated with semantic role labels. The semantic labels were used to specialize the vocabulary of non terminals of the shallow parsing. | constrained submission / predicate arguments were not used to improve shallow parsing in this run |
30.3 |
30.3 |
29.3 |
29.3 |
0.584 |
| UoW-c (Details) | wilkeraziz University of Wolverhampton |
Tree-to-string model trained using the entire Europarl-v6. The English side was shallow parsed and the chunks were relaxed following ideas of syntax-augmented machine translation (Zollmann and Venugopal). | constrained submission |
30.3 |
30.3 |
29.3 |
29.3 |
0.584 |
| PROMT DeepHybrid (Details) | Alexander Molchanov PROMT LLC |
PROMT DeepHybrid is the first version of PROMT hybrid system. The system is based on PROMT RBMT-engine combined with statistical approach. Trained on Europarliament + NewsCommentary. | second run (Europarliament + NewsComm), different training parameters |
29.6 |
29.6 |
28.6 |
28.6 |
0.592 |
| UoW (Details) | wilkeraziz University of Wolverhampton |
Tree-to-string model trained using the entire Europarl-v6. The English side was shallow parsed and annotated with semantic role labels. The semantic labels were used to specialize the vocabulary of non terminals of the shallow parsing. | [primary], constrained submission |
29.0 |
29.0 |
28.1 |
28.1 |
0.601 |
| PROMT DeepHybrid (Details) | Alexander Molchanov PROMT LLC |
PROMT DeepHybrid is the first version of PROMT hybrid system. The system is based on PROMT RBMT-engine combined with statistical approach. Trained on Europarliament + NewsCommentary. | PROMT DeepHybrid engine (hybrid approach based on PROMT RBMT-engine) |
failed |
28.7 |
failed |
27.7 |
0.603 |
| TransductiveMoses (Details) | ergunbicici Koc University |
Moses system with training set selected using a transductive learning approach. | Secondary |
28.4 |
28.4 |
25.5 |
25.5 |
0.608 |
| TransductiveMoses (Details) | ergunbicici Koc University |
Moses system with training set selected using a transductive learning approach. | Primary |
28.3 |
28.3 |
25.5 |
25.5 |
0.606 |
| uk-dan (Details) | zeman Charles University in Prague, ÚFAL |
en-es contrastive stc |
27.6 |
27.6 |
26.5 |
26.5 |
0.621 |
|
| uk-dan (Details) | zeman Charles University in Prague, ÚFAL |
en-es primary |
27.4 |
27.4 |
22.1 |
22.1 |
0.621 |
|
| GTH-UPM Moses with corpus selection (Details) | lapiz Speech Technology Group at UPM |
PRIMARY Moses with train corpus selection - Moses based system - Reordering: msd-bidirectional-fe - Max phrase size 7 - Train corpus selection based on similarity with the source language test - Translation Model: 150.000 best sentences from the global training set (Europarl+UNO+New commentaries) - Languge Model: 1700000 best sentences from the global training set - English corpus: de-contraction “havn’t” -> “have not” - Punctuations marks not included in the training corpus - Post-processing: Numbers and dates formatting - Background vocabulary for no translated words | Considering EU+UNO+NC corpus |
21.7 |
21.7 |
20.9 |
20.9 |
0.692 |
| uedin-wmt11-en-es (Details) | Edinburgh University of Edinburgh |
failed |
failed |