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	ORGANIZER	JPCja-zh	2016/07/13 15:11:50	966	-	30.60	-	-	32.03	31.25	-	0.00	0.00	0.00	SMT	No	Phrase-based SMT
2	ORGANIZER	JPCja-zh	2016/07/13 15:16:28	967	-	30.26	-	-	31.57	30.91	-	0.00	0.00	0.00	SMT	No	Hierarchical Phrase-based SMT
3	ORGANIZER	JPCja-zh	2016/07/13 15:31:58	968	-	31.05	-	-	32.35	31.70	-	0.00	0.00	0.00	SMT	No	String-to-Tree SMT
4	ORGANIZER	JPCja-zh	2016/07/26 10:35:57	1038	-	23.02	-	-	23.57	23.29	-	0.00	0.00	0.00	Other	Yes	Online A (2016)
5	ORGANIZER	JPCja-zh	2016/08/01 18:33:20	1069	-	9.42	-	-	9.59	8.79	-	0.00	0.00	0.00	Other	Yes	Online B (2016)
6	NICT-2	JPCja-zh	2016/08/04 17:37:25	1081	-	33.35	-	-	34.64	33.81	-	0.00	0.00	0.00	SMT	No	Phrase-based SMT with Preordering + Domain Adaptation
7	NICT-2	JPCja-zh	2016/08/05 18:13:16	1106	-	33.40	-	-	34.64	33.83	-	0.00	0.00	0.00	SMT	Yes	Phrase-based SMT with Preordering + Domain Adaptation (JPC and ASPEC)
8	ORGANIZER	JPCja-zh	2016/08/08 17:57:51	1118	-	12.35	-	-	13.72	13.17	-	0.00	0.00	0.00	Other	Yes	RBMT C (2016)
9	bjtu_nlp	JPCja-zh	2016/08/16 12:45:01	1150	-	31.49	-	-	32.79	32.51	-	0.00	0.00	0.00	NMT	No	RNN Encoder-Decoder with attention mechanism, single model
10	Sense	JPCja-zh	2016/08/29 01:08:54	1282	-	29.87	-	-	32.11	30.75	-	0.00	0.00	0.00	SMT	No	Clustercat-C10-PBMT
11	Sense	JPCja-zh	2016/08/29 09:52:57	1283	-	29.59	-	-	31.84	30.46	-	0.00	0.00	0.00	SMT	No	Baseline-C10-PBMT
12	Sense	JPCja-zh	2016/08/29 23:10:28	1293	-	29.59	-	-	31.88	30.44	-	0.00	0.00	0.00	SMT	No	Baseline-C50-PBMT
13	Sense	JPCja-zh	2016/08/30 08:15:19	1295	-	29.44	-	-	31.71	30.36	-	0.00	0.00	0.00	SMT	No	Clustercat-C50-PBMT
14	ORGANIZER	JPCja-zh	2016/11/16 11:16:25	1340	-	33.04	-	-	33.92	33.34	-	0.00	0.00	0.00	NMT	Yes	Online A (2016/11/14)
15	u-tkb	JPCja-zh	2017/07/26 12:33:41	1465	-	31.80	-	-	33.19	32.83	-	0.00	0.00	0.00	NMT	No	NMT with SMT phrase translation (phrase extraction with branching entropy; attention over bidirectional LSTMs; by Harvard NMT)
16	ORGANIZER	JPCja-zh	2018/08/15 18:14:01	1960	-	39.07	-	-	40.32	39.75	-	-	0.00	0.00	NMT	No	NMT with Attention
17	USTC	JPCja-zh	2018/08/31 16:54:46	2202	-	39.71	-	-	40.54	40.05	-	-	0.00	0.00	NMT	No	tensor2tensor, 4 model average, r2l rerank
18	ryan	JPCja-zh	2019/07/25 22:04:25	2950	-	41.22	-	-	42.37	41.99	-	-	-	-	NMT	No	Base Transformer
19	sarah	JPCja-zh	2019/07/26 11:39:52	2982	-	42.13	-	-	43.27	42.98	-	-	-	-	NMT	No	Transformer, ensemble of 4 models
20	KNU_Hyundai	JPCja-zh	2019/07/27 08:37:11	3162	-	43.19	-	-	44.56	43.86	-	-	-	-	NMT	Yes	Transformer(base) + Used ASPEC corpus with relative position, bt, r2l rerank, 4-model ensemble(9-check point)
21	goku20	JPCja-zh	2020/09/21 11:55:29	4088	-	39.49	-	-	40.61	40.14	-	-	-	-	NMT	No	mBART pre-training transformer, single model
22	goku20	JPCja-zh	2020/09/22 00:06:06	4106	-	40.47	-	-	41.68	41.22	-	-	-	-	NMT	No	mBART pre-training transformer, ensemble of 3 models
23	tpt_wat	JPCja-zh	2021/04/27 01:49:08	5694	-	37.83	-	-	39.17	38.89	-	-	-	-	NMT	No	Base Transformer model with separated vocabs 8k size
24	Bering Lab	JPCja-zh	2021/04/30 14:56:11	5853	-	43.48	-	-	44.67	44.24	-	-	-	-	NMT	Yes	Transformer Ensemble with additional crawled parallel corpus
25	sakura	JPCja-zh	2024/08/08 19:34:30	7263	-	40.80	-	-	41.70	41.30	-	-	-	-	NMT	No	LLM: Rakuten/RakutenAI-7B-chat Fine-Tuned with JPC Corpus in six direction (En-Ja, Ja-En, Ko-Ja, Ja-Ko, Zh-Ja, Ja-Zh) - Best
26	sakura	JPCja-zh	2024/08/09 00:26:12	7297	-	40.70	-	-	41.50	41.20	-	-	-	-	NMT	No	LLM: Rakuten/RakutenAI-7B-chat Fine-Tuned with JPC Corpus in six direction (En-Ja, Ja-En, Ko-Ja, Ja-Ko, Zh-Ja, Ja-Zh)

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	ORGANIZER	JPCja-zh	2016/07/13 15:11:50	966	-	0.787321	-	-	0.797888	0.794388	-	0.000000	0.000000	0.000000	SMT	No	Phrase-based SMT
2	ORGANIZER	JPCja-zh	2016/07/13 15:16:28	967	-	0.788415	-	-	0.799118	0.796685	-	0.000000	0.000000	0.000000	SMT	No	Hierarchical Phrase-based SMT
3	ORGANIZER	JPCja-zh	2016/07/13 15:31:58	968	-	0.793846	-	-	0.802805	0.800848	-	0.000000	0.000000	0.000000	SMT	No	String-to-Tree SMT
4	ORGANIZER	JPCja-zh	2016/07/26 10:35:57	1038	-	0.754241	-	-	0.760672	0.760148	-	0.000000	0.000000	0.000000	Other	Yes	Online A (2016)
5	ORGANIZER	JPCja-zh	2016/08/01 18:33:20	1069	-	0.642026	-	-	0.651070	0.643520	-	0.000000	0.000000	0.000000	Other	Yes	Online B (2016)
6	NICT-2	JPCja-zh	2016/08/04 17:37:25	1081	-	0.808513	-	-	0.817996	0.815322	-	0.000000	0.000000	0.000000	SMT	No	Phrase-based SMT with Preordering + Domain Adaptation
7	NICT-2	JPCja-zh	2016/08/05 18:13:16	1106	-	0.811788	-	-	0.820320	0.818701	-	0.000000	0.000000	0.000000	SMT	Yes	Phrase-based SMT with Preordering + Domain Adaptation (JPC and ASPEC)
8	ORGANIZER	JPCja-zh	2016/08/08 17:57:51	1118	-	0.688240	-	-	0.708681	0.700210	-	0.000000	0.000000	0.000000	Other	Yes	RBMT C (2016)
9	bjtu_nlp	JPCja-zh	2016/08/16 12:45:01	1150	-	0.816577	-	-	0.822978	0.820820	-	0.000000	0.000000	0.000000	NMT	No	RNN Encoder-Decoder with attention mechanism, single model
10	Sense	JPCja-zh	2016/08/29 01:08:54	1282	-	0.791760	-	-	0.803288	0.798268	-	0.000000	0.000000	0.000000	SMT	No	Clustercat-C10-PBMT
11	Sense	JPCja-zh	2016/08/29 09:52:57	1283	-	0.790983	-	-	0.803400	0.797822	-	0.000000	0.000000	0.000000	SMT	No	Baseline-C10-PBMT
12	Sense	JPCja-zh	2016/08/29 23:10:28	1293	-	0.788582	-	-	0.801382	0.796637	-	0.000000	0.000000	0.000000	SMT	No	Baseline-C50-PBMT
13	Sense	JPCja-zh	2016/08/30 08:15:19	1295	-	0.789417	-	-	0.802123	0.796346	-	0.000000	0.000000	0.000000	SMT	No	Clustercat-C50-PBMT
14	ORGANIZER	JPCja-zh	2016/11/16 11:16:25	1340	-	0.824829	-	-	0.829122	0.829067	-	0.000000	0.000000	0.000000	NMT	Yes	Online A (2016/11/14)
15	u-tkb	JPCja-zh	2017/07/26 12:33:41	1465	-	0.819791	-	-	0.826055	0.825025	-	0.000000	0.000000	0.000000	NMT	No	NMT with SMT phrase translation (phrase extraction with branching entropy; attention over bidirectional LSTMs; by Harvard NMT)
16	ORGANIZER	JPCja-zh	2018/08/15 18:14:01	1960	-	0.847112	-	-	0.850851	0.850913	-	-	0.000000	0.000000	NMT	No	NMT with Attention
17	USTC	JPCja-zh	2018/08/31 16:54:46	2202	-	0.846472	-	-	0.850750	0.849818	-	-	0.000000	0.000000	NMT	No	tensor2tensor, 4 model average, r2l rerank
18	ryan	JPCja-zh	2019/07/25 22:04:25	2950	-	0.852969	-	-	0.858204	0.857643	-	-	-	-	NMT	No	Base Transformer
19	sarah	JPCja-zh	2019/07/26 11:39:52	2982	-	0.853579	-	-	0.857805	0.857191	-	-	-	-	NMT	No	Transformer, ensemble of 4 models
20	KNU_Hyundai	JPCja-zh	2019/07/27 08:37:11	3162	-	0.860783	-	-	0.865852	0.864763	-	-	-	-	NMT	Yes	Transformer(base) + Used ASPEC corpus with relative position, bt, r2l rerank, 4-model ensemble(9-check point)
21	goku20	JPCja-zh	2020/09/21 11:55:29	4088	-	0.845571	-	-	0.851173	0.850139	-	-	-	-	NMT	No	mBART pre-training transformer, single model
22	goku20	JPCja-zh	2020/09/22 00:06:06	4106	-	0.848852	-	-	0.855067	0.854644	-	-	-	-	NMT	No	mBART pre-training transformer, ensemble of 3 models
23	tpt_wat	JPCja-zh	2021/04/27 01:49:08	5694	-	0.841632	-	-	0.845360	0.844727	-	-	-	-	NMT	No	Base Transformer model with separated vocabs 8k size
24	Bering Lab	JPCja-zh	2021/04/30 14:56:11	5853	-	0.866184	-	-	0.871092	0.869849	-	-	-	-	NMT	Yes	Transformer Ensemble with additional crawled parallel corpus
25	sakura	JPCja-zh	2024/08/08 19:34:30	7263	-	0.856065	-	-	0.859730	0.860075	-	-	-	-	NMT	No	LLM: Rakuten/RakutenAI-7B-chat Fine-Tuned with JPC Corpus in six direction (En-Ja, Ja-En, Ko-Ja, Ja-Ko, Zh-Ja, Ja-Zh) - Best
26	sakura	JPCja-zh	2024/08/09 00:26:12	7297	-	0.855961	-	-	0.859908	0.860476	-	-	-	-	NMT	No	LLM: Rakuten/RakutenAI-7B-chat Fine-Tuned with JPC Corpus in six direction (En-Ja, Ja-En, Ko-Ja, Ja-Ko, Zh-Ja, Ja-Zh)

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	ORGANIZER	JPCja-zh	2016/07/13 15:11:50	966	-	0.710940	-	-	0.710940	0.710940	-	0.000000	0.000000	0.000000	SMT	No	Phrase-based SMT
2	ORGANIZER	JPCja-zh	2016/07/13 15:16:28	967	-	0.718360	-	-	0.718360	0.718360	-	0.000000	0.000000	0.000000	SMT	No	Hierarchical Phrase-based SMT
3	ORGANIZER	JPCja-zh	2016/07/13 15:31:58	968	-	0.720030	-	-	0.720030	0.720030	-	0.000000	0.000000	0.000000	SMT	No	String-to-Tree SMT
4	ORGANIZER	JPCja-zh	2016/07/26 10:35:57	1038	-	0.702350	-	-	0.702350	0.702350	-	0.000000	0.000000	0.000000	Other	Yes	Online A (2016)
5	ORGANIZER	JPCja-zh	2016/08/01 18:33:20	1069	-	0.527180	-	-	0.527180	0.527180	-	0.000000	0.000000	0.000000	Other	Yes	Online B (2016)
6	NICT-2	JPCja-zh	2016/08/04 17:37:25	1081	-	0.723270	-	-	0.723270	0.723270	-	0.000000	0.000000	0.000000	SMT	No	Phrase-based SMT with Preordering + Domain Adaptation
7	NICT-2	JPCja-zh	2016/08/05 18:13:16	1106	-	0.731520	-	-	0.731520	0.731520	-	0.000000	0.000000	0.000000	SMT	Yes	Phrase-based SMT with Preordering + Domain Adaptation (JPC and ASPEC)
8	ORGANIZER	JPCja-zh	2016/08/08 17:57:51	1118	-	0.475430	-	-	0.475430	0.475430	-	0.000000	0.000000	0.000000	Other	Yes	RBMT C (2016)
9	bjtu_nlp	JPCja-zh	2016/08/16 12:45:01	1150	-	0.701490	-	-	0.701490	0.701490	-	0.000000	0.000000	0.000000	NMT	No	RNN Encoder-Decoder with attention mechanism, single model
10	Sense	JPCja-zh	2016/08/29 01:08:54	1282	-	0.709390	-	-	0.709390	0.709390	-	0.000000	0.000000	0.000000	SMT	No	Clustercat-C10-PBMT
11	Sense	JPCja-zh	2016/08/29 09:52:57	1283	-	0.707590	-	-	0.707590	0.707590	-	0.000000	0.000000	0.000000	SMT	No	Baseline-C10-PBMT
12	Sense	JPCja-zh	2016/08/29 23:10:28	1293	-	0.710030	-	-	0.710030	0.710030	-	0.000000	0.000000	0.000000	SMT	No	Baseline-C50-PBMT
13	Sense	JPCja-zh	2016/08/30 08:15:19	1295	-	0.710150	-	-	0.710150	0.710150	-	0.000000	0.000000	0.000000	SMT	No	Clustercat-C50-PBMT
14	ORGANIZER	JPCja-zh	2016/11/16 11:16:25	1340	-	0.735470	-	-	0.735470	0.735470	-	0.000000	0.000000	0.000000	NMT	Yes	Online A (2016/11/14)
15	u-tkb	JPCja-zh	2017/07/26 12:33:41	1465	-	0.706720	-	-	0.706720	0.706720	-	0.000000	0.000000	0.000000	NMT	No	NMT with SMT phrase translation (phrase extraction with branching entropy; attention over bidirectional LSTMs; by Harvard NMT)
16	ORGANIZER	JPCja-zh	2018/08/15 18:14:01	1960	-	0.752360	-	-	0.752360	0.752360	-	-	0.000000	0.000000	NMT	No	NMT with Attention
17	USTC	JPCja-zh	2018/08/31 16:54:46	2202	-	0.757690	-	-	0.757690	0.757690	-	-	0.000000	0.000000	NMT	No	tensor2tensor, 4 model average, r2l rerank
18	ryan	JPCja-zh	2019/07/25 22:04:25	2950	-	0.000000	-	-	0.000000	0.000000	-	-	-	-	NMT	No	Base Transformer
19	sarah	JPCja-zh	2019/07/26 11:39:52	2982	-	0.000000	-	-	0.000000	0.000000	-	-	-	-	NMT	No	Transformer, ensemble of 4 models
20	KNU_Hyundai	JPCja-zh	2019/07/27 08:37:11	3162	-	0.000000	-	-	0.000000	0.000000	-	-	-	-	NMT	Yes	Transformer(base) + Used ASPEC corpus with relative position, bt, r2l rerank, 4-model ensemble(9-check point)
21	goku20	JPCja-zh	2020/09/21 11:55:29	4088	-	0.000000	-	-	0.000000	0.000000	-	-	-	-	NMT	No	mBART pre-training transformer, single model
22	goku20	JPCja-zh	2020/09/22 00:06:06	4106	-	0.000000	-	-	0.000000	0.000000	-	-	-	-	NMT	No	mBART pre-training transformer, ensemble of 3 models
23	tpt_wat	JPCja-zh	2021/04/27 01:49:08	5694	-	0.885218	-	-	0.885218	0.885218	-	-	-	-	NMT	No	Base Transformer model with separated vocabs 8k size
24	Bering Lab	JPCja-zh	2021/04/30 14:56:11	5853	-	0.899221	-	-	0.899221	0.899221	-	-	-	-	NMT	Yes	Transformer Ensemble with additional crawled parallel corpus
25	sakura	JPCja-zh	2024/08/08 19:34:30	7263	-	0.000000	-	-	0.000000	0.000000	-	-	-	-	NMT	No	LLM: Rakuten/RakutenAI-7B-chat Fine-Tuned with JPC Corpus in six direction (En-Ja, Ja-En, Ko-Ja, Ja-Ko, Zh-Ja, Ja-Zh) - Best
26	sakura	JPCja-zh	2024/08/09 00:26:12	7297	-	0.000000	-	-	0.000000	0.000000	-	-	-	-	NMT	No	LLM: Rakuten/RakutenAI-7B-chat Fine-Tuned with JPC Corpus in six direction (En-Ja, Ja-En, Ko-Ja, Ja-Ko, Zh-Ja, Ja-Zh)

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)

#	Team	Task	Date/Time	DataID	HUMAN	Method	Other Resources	System Description
1	ORGANIZER	JPCja-zh	2016/11/16 11:16:25	1340	32.500	NMT	Yes	Online A (2016/11/14)
2	NICT-2	JPCja-zh	2016/08/05 18:13:16	1106	14.000	SMT	Yes	Phrase-based SMT with Preordering + Domain Adaptation (JPC and ASPEC)
3	ORGANIZER	JPCja-zh	2016/07/13 15:16:28	967	4.750	SMT	No	Hierarchical Phrase-based SMT
4	ORGANIZER	JPCja-zh	2016/07/13 15:31:58	968	4.250	SMT	No	String-to-Tree SMT
5	bjtu_nlp	JPCja-zh	2016/08/16 12:45:01	1150	-1.000	NMT	No	RNN Encoder-Decoder with attention mechanism, single model
6	NICT-2	JPCja-zh	2016/08/04 17:37:25	1081	-11.000	SMT	No	Phrase-based SMT with Preordering + Domain Adaptation
7	ORGANIZER	JPCja-zh	2016/07/26 10:35:57	1038	-23.000	Other	Yes	Online A (2016)
8	ORGANIZER	JPCja-zh	2016/08/08 17:57:51	1118	-41.250	Other	Yes	RBMT C (2016)

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