WAT

The Workshop on Asian Translation

Evaluation Results

BLEU

#	Team	Task	Date/Time	DataID	BLEU										Method	Other Resources	System Description
#	Team	Task	Date/Time	DataID	juman	kytea	mecab	moses- tokenizer	stanford- segmenter- ctb	stanford- segmenter- pku	indic- tokenizer	unuse	myseg	kmseg	Method	Other Resources	System Description
1	cvit	HINDENhi-en	2018/09/14 13:21:46	2331	-	-	-	20.63	-	-	-	-	0.00	0.00	NMT	Yes	ConvS2S Model Uses External Data
2	CUNI	HINDENhi-en	2018/09/15 03:10:30	2381	-	-	-	17.80	-	-	-	-	0.00	0.00	NMT	No	Transformer big, transfer learning from CS-EN 1M steps, only original HI-EN, beam=8; alpha=0.8; averaging of last 8 models after 230k steps
3	cvit	HINDENhi-en	2018/11/06 15:51:54	2563	-	-	-	21.04	-	-	-	-	0.00	0.00	NMT	Yes
4	ORGANIZER	HINDENhi-en	2018/11/13 14:57:12	2567	-	-	-	15.44	-	-	-	-	0.00	0.00	NMT	No	NMT with Attention
5	cvit	HINDENhi-en	2019/03/15 01:23:17	2643	-	-	-	22.62	-	-	-	-	0.00	0.00	NMT	Yes	massive-multi
6	cvit	HINDENhi-en	2019/03/15 01:33:22	2645	-	-	-	20.66	-	-	-	-	0.00	0.00	NMT	Yes	massive-multi + ft
7	cvit	HINDENhi-en	2019/03/22 05:52:47	2658	-	-	-	10.76	-	-	-	-	0.00	0.00	NMT	Yes	many to en (Transformer model) trained on WAT2018 data. Detokenized!
8	cvit	HINDENhi-en	2019/05/27 16:04:36	2681	-	-	-	22.91	-	-	-	-	0.00	0.00	NMT	Yes	massive-multi + bt
9	NICT-5	HINDENhi-en	2019/07/23 17:36:36	2865	-	-	-	19.06	-	-	-	-	-	-	NMT	No	HiEn and TaEn mixed training NMT model. Transformer on t2t (Hi-En is external data)
10	LTRC-MT	HINDENhi-en	2019/07/27 04:04:09	3117	-	-	-	16.32	-	-	-	-	-	-	NMT	No	Transformer Baseline, Only IIT-B data
11	LTRC-MT	HINDENhi-en	2019/07/27 04:49:36	3119	-	-	-	18.64	-	-	-	-	-	-	NMT	No	Transformer Model with Backtranslation
12	LTRC-MT	HINDENhi-en	2019/07/27 05:34:14	3121	-	-	-	17.44	-	-	-	-	-	-	NMT	No	LSTM with attention, Backtranslation, Reinforcement Learning for 1 epoch
13	LTRC-MT	HINDENhi-en	2019/07/27 05:58:33	3124	-	-	-	17.07	-	-	-	-	-	-	NMT	No	LSTM with global attention & Backtranslation
14	cvit	HINDENhi-en	2020/06/10 15:37:12	3418	-	-	-	23.83	-	-	-	-	-	-	NMT	Yes	XX-to-EN model, uses PIB-V0 dataset
15	cvit	HINDENhi-en	2020/07/06 02:29:11	3419	-	-	-	24.65	-	-	-	-	-	-	NMT	Yes	XX-to-EN Model, uses PIB-V1 Data
16	cvit	HINDENhi-en	2020/07/06 06:22:02	3422	-	-	-	21.94	-	-	-	-	-	-	NMT	Yes	Multilingual model, mm-all-iter0
17	cvit	HINDENhi-en	2020/07/06 06:38:14	3423	-	-	-	22.48	-	-	-	-	-	-	NMT	Yes	Multilingual Model, Uses PIB-V0 data. (mm-all-iter1)
18	cvit	HINDENhi-en	2020/07/10 04:28:17	3434	-	-	-	24.82	-	-	-	-	-	-	NMT	Yes	xx-to-en model uses PIB-v2 data
19	cvit	HINDENhi-en	2020/07/20 20:38:09	3441	-	-	-	24.85	-	-	-	-	-	-	NMT	Yes	xx-en model, uses PIB-v2 data
20	cvit	HINDENhi-en	2020/08/18 05:27:08	3446	-	-	-	25.26	-	-	-	-	-	-	NMT	Yes
21	WT	HINDENhi-en	2020/09/03 18:12:32	3638	-	-	-	29.59	-	-	-	-	-	-	NMT	No	Used 5M Back translation news crawl data to train. Method: Transformer NMT; Preprocessing: 1. Removed mixed language sentences 2. moses tokeniser for English and for Hindi indicnlp normaliser and toke
22	ORGANIZER	HINDENhi-en	2016/07/26 10:04:53	1031	-	-	-	21.37	-	-	-	0.00	0.00	0.00	Other	Yes	Online A (2016)
23	ORGANIZER	HINDENhi-en	2016/07/26 13:25:18	1048	-	-	-	15.58	-	-	-	0.00	0.00	0.00	Other	Yes	Online B (2016)
24	ORGANIZER	HINDENhi-en	2016/07/26 15:44:20	1054	-	-	-	10.32	-	-	-	0.00	0.00	0.00	SMT	No	Phrase-based SMT
25	IITP-MT	HINDENhi-en	2016/08/29 15:10:41	1289	-	-	-	9.62	-	-	-	0.00	0.00	0.00	SMT	No	Hierarchical SMT
26	XMUNLP	HINDENhi-en	2017/07/24 08:47:29	1427	-	-	-	13.30	-	-	-	0.00	0.00	0.00	NMT	No	single nmt model
27	XMUNLP	HINDENhi-en	2017/07/26 22:54:46	1488	-	-	-	20.61	-	-	-	0.00	0.00	0.00	NMT	No	single nmt model + monolingual data
28	XMUNLP	HINDENhi-en	2017/07/27 23:00:46	1511	-	-	-	22.44	-	-	-	0.00	0.00	0.00	NMT	No	ensemble of 4 nmt models + monolingual data
29	IITB-MTG	HINDENhi-en	2017/08/01 15:10:09	1726	-	-	-	11.55	-	-	-	0.00	0.00	0.00	NMT	No	NMT with ensemble (last 3 + best validation)

Notice:

This table is sorted by the leftmost segmenters. You can change the segmenter used to sort by clicking each segmenter link.

RIBES

#	Team	Task	Date/Time	DataID	RIBES										Method	Other Resources	System Description
#	Team	Task	Date/Time	DataID	juman	kytea	mecab	moses- tokenizer	stanford- segmenter- ctb	stanford- segmenter- pku	indic- tokenizer	unuse	myseg	kmseg	Method	Other Resources	System Description
1	cvit	HINDENhi-en	2018/09/14 13:21:46	2331	-	-	-	0.751883	-	-	-	-	0.000000	0.000000	NMT	Yes	ConvS2S Model Uses External Data
2	CUNI	HINDENhi-en	2018/09/15 03:10:30	2381	-	-	-	0.731727	-	-	-	-	0.000000	0.000000	NMT	No	Transformer big, transfer learning from CS-EN 1M steps, only original HI-EN, beam=8; alpha=0.8; averaging of last 8 models after 230k steps
3	cvit	HINDENhi-en	2018/11/06 15:51:54	2563	-	-	-	0.755941	-	-	-	-	0.000000	0.000000	NMT	Yes
4	ORGANIZER	HINDENhi-en	2018/11/13 14:57:12	2567	-	-	-	0.718751	-	-	-	-	0.000000	0.000000	NMT	No	NMT with Attention
5	cvit	HINDENhi-en	2019/03/15 01:23:17	2643	-	-	-	0.766180	-	-	-	-	0.000000	0.000000	NMT	Yes	massive-multi
6	cvit	HINDENhi-en	2019/03/15 01:33:22	2645	-	-	-	0.758910	-	-	-	-	0.000000	0.000000	NMT	Yes	massive-multi + ft
7	cvit	HINDENhi-en	2019/03/22 05:52:47	2658	-	-	-	0.667353	-	-	-	-	0.000000	0.000000	NMT	Yes	many to en (Transformer model) trained on WAT2018 data. Detokenized!
8	cvit	HINDENhi-en	2019/05/27 16:04:36	2681	-	-	-	0.768324	-	-	-	-	0.000000	0.000000	NMT	Yes	massive-multi + bt
9	NICT-5	HINDENhi-en	2019/07/23 17:36:36	2865	-	-	-	0.741197	-	-	-	-	-	-	NMT	No	HiEn and TaEn mixed training NMT model. Transformer on t2t (Hi-En is external data)
10	LTRC-MT	HINDENhi-en	2019/07/27 04:04:09	3117	-	-	-	0.729072	-	-	-	-	-	-	NMT	No	Transformer Baseline, Only IIT-B data
11	LTRC-MT	HINDENhi-en	2019/07/27 04:49:36	3119	-	-	-	0.735358	-	-	-	-	-	-	NMT	No	Transformer Model with Backtranslation
12	LTRC-MT	HINDENhi-en	2019/07/27 05:34:14	3121	-	-	-	0.735357	-	-	-	-	-	-	NMT	No	LSTM with attention, Backtranslation, Reinforcement Learning for 1 epoch
13	LTRC-MT	HINDENhi-en	2019/07/27 05:58:33	3124	-	-	-	0.729059	-	-	-	-	-	-	NMT	No	LSTM with global attention & Backtranslation
14	cvit	HINDENhi-en	2020/06/10 15:37:12	3418	-	-	-	0.770450	-	-	-	-	-	-	NMT	Yes	XX-to-EN model, uses PIB-V0 dataset
15	cvit	HINDENhi-en	2020/07/06 02:29:11	3419	-	-	-	0.774354	-	-	-	-	-	-	NMT	Yes	XX-to-EN Model, uses PIB-V1 Data
16	cvit	HINDENhi-en	2020/07/06 06:22:02	3422	-	-	-	0.763418	-	-	-	-	-	-	NMT	Yes	Multilingual model, mm-all-iter0
17	cvit	HINDENhi-en	2020/07/06 06:38:14	3423	-	-	-	0.766637	-	-	-	-	-	-	NMT	Yes	Multilingual Model, Uses PIB-V0 data. (mm-all-iter1)
18	cvit	HINDENhi-en	2020/07/10 04:28:17	3434	-	-	-	0.775515	-	-	-	-	-	-	NMT	Yes	xx-to-en model uses PIB-v2 data
19	cvit	HINDENhi-en	2020/07/20 20:38:09	3441	-	-	-	0.774830	-	-	-	-	-	-	NMT	Yes	xx-en model, uses PIB-v2 data
20	cvit	HINDENhi-en	2020/08/18 05:27:08	3446	-	-	-	0.777445	-	-	-	-	-	-	NMT	Yes
21	WT	HINDENhi-en	2020/09/03 18:12:32	3638	-	-	-	0.792065	-	-	-	-	-	-	NMT	No	Used 5M Back translation news crawl data to train. Method: Transformer NMT; Preprocessing: 1. Removed mixed language sentences 2. moses tokeniser for English and for Hindi indicnlp normaliser and toke
22	ORGANIZER	HINDENhi-en	2016/07/26 10:04:53	1031	-	-	-	0.714537	-	-	-	0.000000	0.000000	0.000000	Other	Yes	Online A (2016)
23	ORGANIZER	HINDENhi-en	2016/07/26 13:25:18	1048	-	-	-	0.683214	-	-	-	0.000000	0.000000	0.000000	Other	Yes	Online B (2016)
24	ORGANIZER	HINDENhi-en	2016/07/26 15:44:20	1054	-	-	-	0.638090	-	-	-	0.000000	0.000000	0.000000	SMT	No	Phrase-based SMT
25	IITP-MT	HINDENhi-en	2016/08/29 15:10:41	1289	-	-	-	0.628666	-	-	-	0.000000	0.000000	0.000000	SMT	No	Hierarchical SMT
26	XMUNLP	HINDENhi-en	2017/07/24 08:47:29	1427	-	-	-	0.697707	-	-	-	0.000000	0.000000	0.000000	NMT	No	single nmt model
27	XMUNLP	HINDENhi-en	2017/07/26 22:54:46	1488	-	-	-	0.743656	-	-	-	0.000000	0.000000	0.000000	NMT	No	single nmt model + monolingual data
28	XMUNLP	HINDENhi-en	2017/07/27 23:00:46	1511	-	-	-	0.750921	-	-	-	0.000000	0.000000	0.000000	NMT	No	ensemble of 4 nmt models + monolingual data
29	IITB-MTG	HINDENhi-en	2017/08/01 15:10:09	1726	-	-	-	0.682902	-	-	-	0.000000	0.000000	0.000000	NMT	No	NMT with ensemble (last 3 + best validation)

Notice:

This table is sorted by the leftmost segmenters. You can change the segmenter used to sort by clicking each segmenter link.

AMFM

#	Team	Task	Date/Time	DataID	AMFM										Method	Other Resources	System Description
#	Team	Task	Date/Time	DataID	unuse	unuse	unuse	unuse	unuse	unuse	unuse	unuse	unuse	unuse	Method	Other Resources	System Description
1	cvit	HINDENhi-en	2018/09/14 13:21:46	2331	-	-	-	0.623240	-	-	-	-	0.000000	0.000000	NMT	Yes	ConvS2S Model Uses External Data
2	CUNI	HINDENhi-en	2018/09/15 03:10:30	2381	-	-	-	0.611090	-	-	-	-	0.000000	0.000000	NMT	No	Transformer big, transfer learning from CS-EN 1M steps, only original HI-EN, beam=8; alpha=0.8; averaging of last 8 models after 230k steps
3	cvit	HINDENhi-en	2018/11/06 15:51:54	2563	-	-	-	0.628600	-	-	-	-	0.000000	0.000000	NMT	Yes
4	ORGANIZER	HINDENhi-en	2018/11/13 14:57:12	2567	-	-	-	0.586360	-	-	-	-	0.000000	0.000000	NMT	No	NMT with Attention
5	cvit	HINDENhi-en	2019/03/15 01:23:17	2643	-	-	-	0.637230	-	-	-	-	0.000000	0.000000	NMT	Yes	massive-multi
6	cvit	HINDENhi-en	2019/03/15 01:33:22	2645	-	-	-	0.631250	-	-	-	-	0.000000	0.000000	NMT	Yes	massive-multi + ft
7	cvit	HINDENhi-en	2019/03/22 05:52:47	2658	-	-	-	0.554700	-	-	-	-	0.000000	0.000000	NMT	Yes	many to en (Transformer model) trained on WAT2018 data. Detokenized!
8	cvit	HINDENhi-en	2019/05/27 16:04:36	2681	-	-	-	0.641730	-	-	-	-	0.000000	0.000000	NMT	Yes	massive-multi + bt
9	NICT-5	HINDENhi-en	2019/07/23 17:36:36	2865	-	-	-	0.566490	-	-	-	-	-	-	NMT	No	HiEn and TaEn mixed training NMT model. Transformer on t2t (Hi-En is external data)
10	LTRC-MT	HINDENhi-en	2019/07/27 04:04:09	3117	-	-	-	0.563590	-	-	-	-	-	-	NMT	No	Transformer Baseline, Only IIT-B data
11	LTRC-MT	HINDENhi-en	2019/07/27 04:49:36	3119	-	-	-	0.594770	-	-	-	-	-	-	NMT	No	Transformer Model with Backtranslation
12	LTRC-MT	HINDENhi-en	2019/07/27 05:34:14	3121	-	-	-	0.594550	-	-	-	-	-	-	NMT	No	LSTM with attention, Backtranslation, Reinforcement Learning for 1 epoch
13	LTRC-MT	HINDENhi-en	2019/07/27 05:58:33	3124	-	-	-	0.587060	-	-	-	-	-	-	NMT	No	LSTM with global attention & Backtranslation
14	cvit	HINDENhi-en	2020/06/10 15:37:12	3418	-	-	-	0.601810	-	-	-	-	-	-	NMT	Yes	XX-to-EN model, uses PIB-V0 dataset
15	cvit	HINDENhi-en	2020/07/06 02:29:11	3419	-	-	-	0.609500	-	-	-	-	-	-	NMT	Yes	XX-to-EN Model, uses PIB-V1 Data
16	cvit	HINDENhi-en	2020/07/06 06:22:02	3422	-	-	-	0.596650	-	-	-	-	-	-	NMT	Yes	Multilingual model, mm-all-iter0
17	cvit	HINDENhi-en	2020/07/06 06:38:14	3423	-	-	-	0.596890	-	-	-	-	-	-	NMT	Yes	Multilingual Model, Uses PIB-V0 data. (mm-all-iter1)
18	cvit	HINDENhi-en	2020/07/10 04:28:17	3434	-	-	-	0.610650	-	-	-	-	-	-	NMT	Yes	xx-to-en model uses PIB-v2 data
19	cvit	HINDENhi-en	2020/07/20 20:38:09	3441	-	-	-	0.610910	-	-	-	-	-	-	NMT	Yes	xx-en model, uses PIB-v2 data
20	cvit	HINDENhi-en	2020/08/18 05:27:08	3446	-	-	-	0.614060	-	-	-	-	-	-	NMT	Yes
21	WT	HINDENhi-en	2020/09/03 18:12:32	3638	-	-	-	0.637410	-	-	-	-	-	-	NMT	No	Used 5M Back translation news crawl data to train. Method: Transformer NMT; Preprocessing: 1. Removed mixed language sentences 2. moses tokeniser for English and for Hindi indicnlp normaliser and toke
22	ORGANIZER	HINDENhi-en	2016/07/26 10:04:53	1031	-	-	-	0.621100	-	-	-	0.000000	0.000000	0.000000	Other	Yes	Online A (2016)
23	ORGANIZER	HINDENhi-en	2016/07/26 13:25:18	1048	-	-	-	0.590520	-	-	-	0.000000	0.000000	0.000000	Other	Yes	Online B (2016)
24	ORGANIZER	HINDENhi-en	2016/07/26 15:44:20	1054	-	-	-	0.574850	-	-	-	0.000000	0.000000	0.000000	SMT	No	Phrase-based SMT
25	IITP-MT	HINDENhi-en	2016/08/29 15:10:41	1289	-	-	-	0.567370	-	-	-	0.000000	0.000000	0.000000	SMT	No	Hierarchical SMT
26	XMUNLP	HINDENhi-en	2017/07/24 08:47:29	1427	-	-	-	0.568010	-	-	-	0.000000	0.000000	0.000000	NMT	No	single nmt model
27	XMUNLP	HINDENhi-en	2017/07/26 22:54:46	1488	-	-	-	0.627190	-	-	-	0.000000	0.000000	0.000000	NMT	No	single nmt model + monolingual data
28	XMUNLP	HINDENhi-en	2017/07/27 23:00:46	1511	-	-	-	0.629530	-	-	-	0.000000	0.000000	0.000000	NMT	No	ensemble of 4 nmt models + monolingual data
29	IITB-MTG	HINDENhi-en	2017/08/01 15:10:09	1726	-	-	-	0.557040	-	-	-	0.000000	0.000000	0.000000	NMT	No	NMT with ensemble (last 3 + best validation)

Notice:

This table is sorted by the leftmost segmenters. You can change the segmenter used to sort by clicking each segmenter link.
Adequacy-Fluency Metrics (AMFM) is a two-dimensional automatic evaluation metric for machine translation, designed to operate at the sentence level. It is based on adequacy and fluency, to decouple semantic and syntactic components of the translation process to provide a balanced view on translation quality.
AMFM is calculated without tokenizers.
The detail of AMFM is shown on the following paper: "Adequacy–Fluency Metrics: Evaluating MT in the Continuous Space Model Framework" [pdf]. Invited Talk in WAT2015 also helps understanding [slide].

HUMAN (WAT2022)

Notice:

HUMAN (WAT2022) is the result of the Pairwise Crowdsourcing Evaluation on WAT2022.
HUMAN (WAT2022) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.

HUMAN (WAT2021)

Notice:

HUMAN (WAT2021) is the result of the Pairwise Crowdsourcing Evaluation on WAT2021.
HUMAN (WAT2021) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.

HUMAN (WAT2020)

Notice:

HUMAN (WAT2020) is the result of the Pairwise Crowdsourcing Evaluation on WAT2020.
HUMAN (WAT2020) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.

HUMAN (WAT2019)

Notice:

HUMAN (WAT2019) is the result of the Pairwise Crowdsourcing Evaluation on WAT2019.
HUMAN (WAT2019) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.

HUMAN (WAT2018)

Notice:

HUMAN (WAT2018) is the result of the Pairwise Crowdsourcing Evaluation on WAT2018.
HUMAN (WAT2018) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.

HUMAN (WAT2017)

Notice:

HUMAN (WAT2017) is the result of the Pairwise Crowdsourcing Evaluation on WAT2017.
HUMAN (WAT2017) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.

HUMAN (WAT2016)

Notice:

HUMAN (WAT2016) is the result of the Pairwise Crowdsourcing Evaluation on WAT2016.
HUMAN (WAT2016) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.

HUMAN (WAT2015)

Notice:

HUMAN (WAT2015) is the result of the Pairwise Crowdsourcing Evaluation on WAT2015.
HUMAN (WAT2015) was evaluated by 5 different workers and the final decision is made by the voting of the judgements.
The detail of the evaluation can be found in the pdf document (PDF file).

HUMAN (WAT2014)

Notice:

HUMAN (WAT2014) is the result of the Pairwise Crowdsourcing Evaluation on WAT2014.
HUMAN (WAT2014) was evaluated by 3 different workers and the final decision is made by the voting of the judgements.
The detail of the evaluation can be found in the pdf document (PDF file).

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

WAT The Workshop on Asian Translation Evaluation Results

BLEU

RIBES

AMFM

HUMAN (WAT2022)

HUMAN (WAT2021)

HUMAN (WAT2020)

HUMAN (WAT2019)

HUMAN (WAT2018)

HUMAN (WAT2017)

HUMAN (WAT2016)

HUMAN (WAT2015)

HUMAN (WAT2014)

EVALUATION RESULTS USAGE POLICY

WAT

The Workshop on Asian Translation

Evaluation Results