NICT_LOGO.JPG KYOTO-U_LOGO.JPG

WAT

The Workshop on Asian Translation
Evaluation Results

[EVALUATION RESULTS TOP] | [BLEU] | [RIBES] | [AMFM] | [HUMAN (WAT2019)] | [HUMAN (WAT2018)] | [HUMAN (WAT2017)] | [HUMAN (WAT2016)] | [HUMAN (WAT2015)] | [HUMAN (WAT2014)] | [EVALUATION RESULTS USAGE POLICY]

BLEU


# Team Task Date/Time DataID BLEU
Method
Other
Resources
System
Description
juman kytea mecab moses-
tokenizer
stanford-
segmenter-
ctb
stanford-
segmenter-
pku
indic-
tokenizer
unuse myseg kmseg
1NHK-NESJIJIja-en2019/07/24 14:03:102883---26.38------NMTYesensemble-10, beamwidth-30 with other resources (including BT)
2sarahJIJIja-en2019/07/22 11:59:112793---21.84------NMTNoTransformer, ensemble of 4 models
3sarahJIJIja-en2019/07/22 16:41:432813---21.34------NMTNoTransformer, single model
4NTTJIJIja-en2017/08/01 02:55:311677---20.90--- 0.00 0.00 0.00NMTNoEnsemble 8 Models: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
5NTTJIJIja-en2017/07/30 10:16:391599---19.44--- 0.00 0.00 0.00NMTNoSingle Model: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
6NICT-2JIJIja-en2017/07/26 13:49:391474---18.19--- 0.00 0.00 0.00NMTNoNMT 16 Ensembles * Bi-directional Reranking
7XMUNLPJIJIja-en2017/07/24 20:42:391442---17.95--- 0.00 0.00 0.00NMTNoensemble of 4 nmt models
8NICT-2JIJIja-en2017/07/26 13:46:111473---16.52--- 0.00 0.00 0.00NMTNoNMT Single Model: BPE35k, Bi-LSTM(500*2) Encoder, LSTM(1000) Left-to-Right Decoder
9ORGANIZERJIJIja-en2018/08/14 11:49:181905---16.48---- 0.00 0.00NMTNoNMT with Attention
10XMUNLPJIJIja-en2017/07/24 09:02:371428---15.77--- 0.00 0.00 0.00NMTNosingle nmt model
11NTTJIJIja-en2017/07/28 19:24:021564---15.77--- 0.00 0.00 0.00NMTYesSingle Model: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 5 w/ length-based reranking / initialized model = ASPEC + JPO
12ORGANIZERJIJIja-en2017/07/19 13:12:431396---15.67--- 0.00 0.00 0.00SMTNoHierarchical Phrase-based SMT
13ORGANIZERJIJIja-en2017/07/19 12:51:031394---15.11--- 0.00 0.00 0.00SMTNoPhrase-based SMT
14ORGANIZERJIJIja-en2017/07/19 13:39:221398---14.54--- 0.00 0.00 0.00SMTNoString-to-Tree SMT
15NHK-NESJIJIja-en2019/07/24 14:07:562884---14.23------NMTYesensemble-10, beamwidth-30 using another domain-tag (including BT)
16CUNIJIJIja-en2017/07/31 22:38:531668---10.67--- 0.00 0.00 0.00SMTNoBahdanau (2014) seq2seq with conditional GRU on byte-pair encoding
17ORGANIZERJIJIja-en2017/07/28 15:24:581523--- 8.19--- 0.00 0.00 0.00NMTNoONLINE-A
18ORGANIZERJIJIja-en2017/07/28 15:25:471524--- 5.17--- 0.00 0.00 0.00NMTNoONLINE-B
19ORGANIZERJIJIja-en2017/07/28 15:27:391526--- 4.67--- 0.00 0.00 0.00RBMTNoRBMT-B
20ORGANIZERJIJIja-en2017/07/28 15:26:561525--- 4.36--- 0.00 0.00 0.00RBMTNoRBMT-A

Notice:

Back to top

RIBES


# Team Task Date/Time DataID RIBES
Method
Other
Resources
System
Description
juman kytea mecab moses-
tokenizer
stanford-
segmenter-
ctb
stanford-
segmenter-
pku
indic-
tokenizer
unuse myseg kmseg
1NHK-NESJIJIja-en2019/07/24 14:03:102883---0.703808------NMTYesensemble-10, beamwidth-30 with other resources (including BT)
2sarahJIJIja-en2019/07/22 16:41:432813---0.676723------NMTNoTransformer, single model
3sarahJIJIja-en2019/07/22 11:59:112793---0.675386------NMTNoTransformer, ensemble of 4 models
4NTTJIJIja-en2017/08/01 02:55:311677---0.648931---0.0000000.0000000.000000NMTNoEnsemble 8 Models: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
5NICT-2JIJIja-en2017/07/26 13:46:111473---0.642379---0.0000000.0000000.000000NMTNoNMT Single Model: BPE35k, Bi-LSTM(500*2) Encoder, LSTM(1000) Left-to-Right Decoder
6ORGANIZERJIJIja-en2018/08/14 11:49:181905---0.640558----0.0000000.000000NMTNoNMT with Attention
7NTTJIJIja-en2017/07/30 10:16:391599---0.638841---0.0000000.0000000.000000NMTNoSingle Model: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
8XMUNLPJIJIja-en2017/07/24 20:42:391442---0.637059---0.0000000.0000000.000000NMTNoensemble of 4 nmt models
9NICT-2JIJIja-en2017/07/26 13:49:391474---0.632638---0.0000000.0000000.000000NMTNoNMT 16 Ensembles * Bi-directional Reranking
10NTTJIJIja-en2017/07/28 19:24:021564---0.627182---0.0000000.0000000.000000NMTYesSingle Model: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 5 w/ length-based reranking / initialized model = ASPEC + JPO
11XMUNLPJIJIja-en2017/07/24 09:02:371428---0.617123---0.0000000.0000000.000000NMTNosingle nmt model
12NHK-NESJIJIja-en2019/07/24 14:07:562884---0.612351------NMTYesensemble-10, beamwidth-30 using another domain-tag (including BT)
13CUNIJIJIja-en2017/07/31 22:38:531668---0.564797---0.0000000.0000000.000000SMTNoBahdanau (2014) seq2seq with conditional GRU on byte-pair encoding
14ORGANIZERJIJIja-en2017/07/19 13:12:431396---0.558225---0.0000000.0000000.000000SMTNoHierarchical Phrase-based SMT
15ORGANIZERJIJIja-en2017/07/19 13:39:221398---0.556728---0.0000000.0000000.000000SMTNoString-to-Tree SMT
16ORGANIZERJIJIja-en2017/07/19 12:51:031394---0.554550---0.0000000.0000000.000000SMTNoPhrase-based SMT
17ORGANIZERJIJIja-en2017/07/28 15:24:581523---0.529844---0.0000000.0000000.000000NMTNoONLINE-A
18ORGANIZERJIJIja-en2017/07/28 15:27:391526---0.475760---0.0000000.0000000.000000RBMTNoRBMT-B
19ORGANIZERJIJIja-en2017/07/28 15:26:561525---0.472312---0.0000000.0000000.000000RBMTNoRBMT-A
20ORGANIZERJIJIja-en2017/07/28 15:25:471524---0.461632---0.0000000.0000000.000000NMTNoONLINE-B

Notice:

Back to top

AMFM


# Team Task Date/Time DataID AMFM
Method
Other
Resources
System
Description
juman kytea mecab moses-
tokenizer
stanford-
segmenter-
ctb
stanford-
segmenter-
pku
indic-
tokenizer
unuse myseg kmseg
1NHK-NESJIJIja-en2019/07/24 14:03:102883---0.554110------NMTYesensemble-10, beamwidth-30 with other resources (including BT)
2sarahJIJIja-en2019/07/22 11:59:112793---0.526530------NMTNoTransformer, ensemble of 4 models
3NHK-NESJIJIja-en2019/07/24 14:07:562884---0.526010------NMTYesensemble-10, beamwidth-30 using another domain-tag (including BT)
4sarahJIJIja-en2019/07/22 16:41:432813---0.524530------NMTNoTransformer, single model
5ORGANIZERJIJIja-en2017/07/19 13:39:221398---0.477170---0.0000000.0000000.000000SMTNoString-to-Tree SMT
6NTTJIJIja-en2017/07/30 10:16:391599---0.476200---0.0000000.0000000.000000NMTNoSingle Model: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
7ORGANIZERJIJIja-en2017/07/19 12:51:031394---0.475740---0.0000000.0000000.000000SMTNoPhrase-based SMT
8NTTJIJIja-en2017/08/01 02:55:311677---0.474360---0.0000000.0000000.000000NMTNoEnsemble 8 Models: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
9NTTJIJIja-en2017/07/28 19:24:021564---0.473970---0.0000000.0000000.000000NMTYesSingle Model: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 5 w/ length-based reranking / initialized model = ASPEC + JPO
10ORGANIZERJIJIja-en2017/07/19 13:12:431396---0.470610---0.0000000.0000000.000000SMTNoHierarchical Phrase-based SMT
11XMUNLPJIJIja-en2017/07/24 20:42:391442---0.465710---0.0000000.0000000.000000NMTNoensemble of 4 nmt models
12XMUNLPJIJIja-en2017/07/24 09:02:371428---0.463550---0.0000000.0000000.000000NMTNosingle nmt model
13ORGANIZERJIJIja-en2018/08/14 11:49:181905---0.459080----0.0000000.000000NMTNoNMT with Attention
14NICT-2JIJIja-en2017/07/26 13:46:111473---0.459000---0.0000000.0000000.000000NMTNoNMT Single Model: BPE35k, Bi-LSTM(500*2) Encoder, LSTM(1000) Left-to-Right Decoder
15NICT-2JIJIja-en2017/07/26 13:49:391474---0.453420---0.0000000.0000000.000000NMTNoNMT 16 Ensembles * Bi-directional Reranking
16ORGANIZERJIJIja-en2017/07/28 15:24:581523---0.450850---0.0000000.0000000.000000NMTNoONLINE-A
17CUNIJIJIja-en2017/07/31 22:38:531668---0.432700---0.0000000.0000000.000000SMTNoBahdanau (2014) seq2seq with conditional GRU on byte-pair encoding
18ORGANIZERJIJIja-en2017/07/28 15:26:561525---0.391050---0.0000000.0000000.000000RBMTNoRBMT-A
19ORGANIZERJIJIja-en2017/07/28 15:27:391526---0.385600---0.0000000.0000000.000000RBMTNoRBMT-B
20ORGANIZERJIJIja-en2017/07/28 15:25:471524---0.367760---0.0000000.0000000.000000NMTNoONLINE-B

Notice:

Back to top

HUMAN (WAT2019)


# Team Task Date/Time DataID HUMAN
Method
Other
Resources
System
Description
1NHK-NESJIJIja-en2019/07/24 14:07:56288489.000NMTYesensemble-10, beamwidth-30 using another domain-tag (including BT)
2NHK-NESJIJIja-en2019/07/24 14:03:10288372.000NMTYesensemble-10, beamwidth-30 with other resources (including BT)
3sarahJIJIja-en2019/07/22 11:59:11279350.750NMTNoTransformer, ensemble of 4 models
4sarahJIJIja-en2019/07/22 16:41:43281344.750NMTNoTransformer, single model

Notice:
Back to top

HUMAN (WAT2018)


# Team Task Date/Time DataID HUMAN
Method
Other
Resources
System
Description

Notice:
Back to top

HUMAN (WAT2017)


# Team Task Date/Time DataID HUMAN
Method
Other
Resources
System
Description
1ORGANIZERJIJIja-en2017/07/28 15:24:58152370.000NMTNoONLINE-A
2ORGANIZERJIJIja-en2017/07/28 15:27:39152651.750RBMTNoRBMT-B
3NTTJIJIja-en2017/07/30 10:16:39159932.000NMTNoSingle Model: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
4NTTJIJIja-en2017/08/01 02:55:31167726.750NMTNoEnsemble 8 Models: joint BPE 16k, BiLSTM Encoder 512*2*2, LtoR LSTM Decoder 512*2, Beam Search 20 w/ length-based reranking
5XMUNLPJIJIja-en2017/07/24 20:42:39144220.750NMTNoensemble of 4 nmt models
6ORGANIZERJIJIja-en2017/07/19 13:12:43139610.250SMTNoHierarchical Phrase-based SMT
7NICT-2JIJIja-en2017/07/26 13:49:3914747.250NMTNoNMT 16 Ensembles * Bi-directional Reranking
8NICT-2JIJIja-en2017/07/26 13:46:1114730.250NMTNoNMT Single Model: BPE35k, Bi-LSTM(500*2) Encoder, LSTM(1000) Left-to-Right Decoder
9CUNIJIJIja-en2017/07/31 22:38:531668-24.000SMTNoBahdanau (2014) seq2seq with conditional GRU on byte-pair encoding

Notice:
Back to top

HUMAN (WAT2016)


# Team Task Date/Time DataID HUMAN
Method
Other
Resources
System
Description

Notice:
Back to top

HUMAN (WAT2015)


# Team Task Date/Time DataID HUMAN
Method
Other
Resources
System
Description

Notice:
Back to top

HUMAN (WAT2014)


# Team Task Date/Time DataID HUMAN
Method
Other
Resources
System
Description

Notice:
Back to top

EVALUATION RESULTS USAGE POLICY

When you use the WAT evaluation results for any purpose such as:
- writing technical papers,
- making presentations about your system,
- advertising your MT system to the customers,
you can use the information about translation directions, scores (including both automatic and human evaluations) and ranks of your system among others. You can also use the scores of the other systems, but you MUST anonymize the other system's names. In addition, you can show the links (URLs) to the WAT evaluation result pages.

NICT (National Institute of Information and Communications Technology)
Kyoto University
Last Modified: 2018-08-02