Interested in Contributing?

Scored Systems

System Submitter System Notes Constraint Run Notes BLEU BLEU-cased TER BEER 2.0 CharactTER
uedin-pbt-wmt15-fi-en  (Details) barry
University of Edinburgh
no Moses, all OPUS, OSM (run 20)

20.7

19.7

0.717

uedin-syntax-fi-en  (Details) Phil Williams
University of Edinburgh
Moses string-to-tree, morphological segmentation with Morfessor 2.0 yes

19.0

17.9

0.738

UU-fien-unconstrained  (Details) jorgtied
University of Helsinki
no factored with morphology and OPUS data

19.3

17.8

0.728

abumatran-fien-combo  (Details) atoral
Dublin City University
combination of unsegmented and segmented models (rule-based and unsupervised) yes

18.7

17.6

0.727

PROMT SMT  (Details) Alex Molchanov
PROMT LLC
Moses phrase-based, built on Opus data + company private data. 3-gram kenLM built on news 2014-2015. Snowball stemmer (NLTK), compound splitting for OOVs. no

18.6

17.6

0.742

UU-fien-unconstrained  (Details) jorgtied
University of Helsinki
no phrase-based baseline with OPUS data

18.9

17.5

0.737

UU-fien  (Details) jorgtied
University of Helsinki
phrase-based system yes factored with morphology

17.9

16.4

0.749

abumatran-fien  (Details) rrubino
Saarland University & DFKI
PB-SMT, OSM, BiNLM, 3 reordering models yes

16.7

15.9

0.764

UIUC fi-en omorfi  (Details) dowobeha
University of Illinois
back-off to morphemes from omorfi for OOVs yes

16.5

15.7

0.764

ParFDA5-DCU  (Details) bicici
Parallel Feature Decay Algorithms (ParFDA5) Moses RELEASE-3.0 phrase based SMT results. yes fi-en (after the deadline).

16.2

15.4

0.775

abumatran-fien-hfstmorph  (Details) rrubino
Saarland University & DFKI
HFST-Morph segmentation, PB-SMT, OSM, BiNLM, 3 reordering models yes

16.3

15.3

0.782

UIUC_prelim  (Details) dowobeha
University of Illinois
Gigaword LM yes

15.7

14.9

0.780

UU-fien  (Details) jorgtied
University of Helsinki
phrase-based system yes simple phrase-based baseline (just for testing the upload)

15.5

14.2

0.780

Neural MT (primary)  (Details) UMontreal-MILA
University of Montreal
Neural machine translation (primary) yes Primary (ensemble)

14.2

13.6

0.789

UIUC fi-en lattice 5-best  (Details) dowobeha
University of Illinois
yes

14.1

13.4

0.806

ParFDA5-DCU  (Details) bicici
Parallel Feature Decay Algorithms (ParFDA5) Moses RELEASE-3.0 phrase based SMT results. yes fi-en, LM is train target.

14.1

13.2

0.798

UoS-nth  (Details) DstKsm
UoS
no

13.4

12.5

0.792

limsi-fien  (Details) MatthieuLabeau
LIMSI-CNRS
no

13.2

12.4

0.823

UoS-nt  (Details) DstKsm
UoS
With partial stemming yes

13.4

12.4

0.792

UoS  (Details) DstKsm
UoS
Europarl and partial newscrawl for larger LM. no

13.4

12.3

0.791

limsi-fien-nm  (Details) MatthieuLabeau
LIMSI-CNRS
yes

14.2

11.9

0.788

Neural MT (single)  (Details) UMontreal-MILA
University of Montreal
Neural machine translation yes Contrastive (single model)

10.5

10.1

0.820

apertium-fin-eng-unconstrained-fien  (Details) tpirinen
Hamburger Zentrum für Sprachkorpora
This is an RBMT with large dictionary and limited amount of rules. The system has been made by "tuning" on dev set, i.e. manually fixing rules optimising for dev set quality no This is an out of the box RBMT with the words of the dev and test sets missing from bilingual dictionary but not monolingual ones added.

7.3

6.9

0.829