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	OdiaGenAI	MMCHTEXT24en-hi	2025/10/23 11:39:05	7485	-	-	-	-	-	-	56.90	-	-	-		Yes	NLLB-3.3B fine-tuned with WAT dataset + 100k pairs from samanantar for Hindi language only.
2	IITP-AI-NLP-ML	MMCHTEXT24en-hi	2025/10/22 22:13:17	7471	-	-	-	-	-	-	56.10	-	-	-	NMT	Yes	Finetuned indicTrans on all 4 languages combined
3	00-7	MMCHTEXT24en-hi	2024/08/09 18:03:29	7313	-	-	-	-	-	-	54.10	-	-	-	NMT	No
4	BLEU Monday	MMCHTEXT24en-hi	2025/10/23 19:25:18	7500	-	-	-	-	-	-	54.00	-	-	-	Other	Yes	LORA on original data
5	BLEU Monday	MMCHTEXT24en-hi	2025/10/23 16:16:00	7490	-	-	-	-	-	-	53.90	-	-	-	Other	Yes	LORA on corrected data
6	ODIAGEN	MMCHTEXT24en-hi	2023/07/03 02:56:51	7088	-	-	-	-	-	-	53.60	-	-	-	NMT	No	Fine-tuning Transformer using NLLB-200 from Facebook
7	ODIAGEN	MMCHTEXT24en-hi	2023/07/06 12:31:36	7110	-	-	-	-	-	-	53.10	-	-	-	NMT	No	Fine-tuning Transformer using NLLB-200 from Facebook
8	Volta	MMCHTEXT24en-hi	2021/05/25 13:55:36	6429	-	-	-	-	-	-	51.66	-	-	-	NMT	Yes	Finetuned mBART (Used IITB corpus for data augmentation)
9	ODIAGEN	MMCHTEXT24en-hi	2024/08/12 23:38:25	7358	-	-	-	-	-	-	44.10	-	-	-	Other	No	LLM based (fine-tuning Mistral7B)
10	nlp_novices	MMCHTEXT24en-hi	2022/07/10 22:51:34	6725	-	-	-	-	-	-	41.80	-	-	-	NMT	Yes	Finetuned Transformers over OPUS Corpora additionally
11	ODIANLP	MMCHTEXT24en-hi	2020/09/14 19:07:05	3713	-	-	-	-	-	-	38.50	-	-	-	NMT	Yes	Transformer model (used IITB as an additional resource for training)
12	CNLP-NITS-PP	MMCHTEXT24en-hi	2022/07/11 12:43:47	6742	-	-	-	-	-	-	37.20	-	-	-	NMT	No	Transliteration-based phrase pairs augmentation in training using BRNN-based NMT
13	CNLP-NITS-PP	MMCHTEXT24en-hi	2021/04/28 00:15:03	5732	-	-	-	-	-	-	37.16	-	-	-	NMT	Yes	Pretrained monolingual data (IITB) and fine-tuned with parallel data (WAT21 train data+ Extracted Phrase pairs from WAT21 train data +IITB train data) in training using BRNN model.
14	DCU_NMT	MMCHTEXT24en-hi	2024/08/11 22:57:15	7349	-	-	-	-	-	-	35.90	-	-	-	NMT	No	En-Hi system trained using backtranslated data from flicker.
15	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:12:48	4188	-	-	-	-	-	-	30.72	-	-	-	NMT	No
16	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:15:08	4189	-	-	-	-	-	-	30.72	-	-	-	NMT	No
17	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:18:17	4190	-	-	-	-	-	-	30.72	-	-	-	NMT	No
18	SILO_NLP	MMCHTEXT24en-hi	2022/07/12 04:15:44	6838	-	-	-	-	-	-	29.60	-	-	-	NMT	No	Fine-tuning with pre-trained mBART-50 model
19	DCU_NMT	MMCHTEXT24en-hi	2024/08/11 23:00:10	7350	-	-	-	-	-	-	29.20	-	-	-	NMT	No	The baseline text-only model trained only on provided data.
20	ORGANIZER	MMCHTEXT24en-hi	2020/08/27 21:08:36	3587	-	-	-	-	-	-	28.35	-	-	-	NMT	No	Transformer base model
21	CNLP-NITS	MMCHTEXT24en-hi	2020/09/18 16:12:32	3898	-	-	-	-	-	-	27.75	-	-	-	NMT	Yes	Pretrained monolingual data (IITB) and fine-tuned with parallel data in training using BRNN model.
22		MMCHTEXT24en-hi	2020/09/19 06:17:42	4031	-	-	-	-	-	-	20.52	-	-	-	NMT	No	For the text-only Eng-Hindi translation task, we use an adaptation of the NMT-Keras code to suit our problem. Here, the focus is on long term translation as well as active learning strategies. The t
23	ODIAGEN	MMCHTEXT24en-hi	2024/08/15 17:16:06	7402	-	-	-	-	-	-	1.10	-	-	-	Other	No	LLM based 8fine-tuned multimodal Llava model for region-specific instruction set in Hindi

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	IITP-AI-NLP-ML	MMCHTEXT24en-hi	2025/10/22 22:13:17	7471	-	-	-	-	-	-	0.870914	-	-	-	NMT	Yes	Finetuned indicTrans on all 4 languages combined
2	OdiaGenAI	MMCHTEXT24en-hi	2025/10/23 11:39:05	7485	-	-	-	-	-	-	0.870254	-	-	-		Yes	NLLB-3.3B fine-tuned with WAT dataset + 100k pairs from samanantar for Hindi language only.
3	BLEU Monday	MMCHTEXT24en-hi	2025/10/23 16:16:00	7490	-	-	-	-	-	-	0.866823	-	-	-	Other	Yes	LORA on corrected data
4	BLEU Monday	MMCHTEXT24en-hi	2025/10/23 19:25:18	7500	-	-	-	-	-	-	0.864790	-	-	-	Other	Yes	LORA on original data
5	00-7	MMCHTEXT24en-hi	2024/08/09 18:03:29	7313	-	-	-	-	-	-	0.858322	-	-	-	NMT	No
6	ODIAGEN	MMCHTEXT24en-hi	2023/07/03 02:56:51	7088	-	-	-	-	-	-	0.858033	-	-	-	NMT	No	Fine-tuning Transformer using NLLB-200 from Facebook
7	Volta	MMCHTEXT24en-hi	2021/05/25 13:55:36	6429	-	-	-	-	-	-	0.855410	-	-	-	NMT	Yes	Finetuned mBART (Used IITB corpus for data augmentation)
8	ODIAGEN	MMCHTEXT24en-hi	2023/07/06 12:31:36	7110	-	-	-	-	-	-	0.854334	-	-	-	NMT	No	Fine-tuning Transformer using NLLB-200 from Facebook
9	ODIAGEN	MMCHTEXT24en-hi	2024/08/12 23:38:25	7358	-	-	-	-	-	-	0.815457	-	-	-	Other	No	LLM based (fine-tuning Mistral7B)
10	nlp_novices	MMCHTEXT24en-hi	2022/07/10 22:51:34	6725	-	-	-	-	-	-	0.812500	-	-	-	NMT	Yes	Finetuned Transformers over OPUS Corpora additionally
11	ODIANLP	MMCHTEXT24en-hi	2020/09/14 19:07:05	3713	-	-	-	-	-	-	0.785252	-	-	-	NMT	Yes	Transformer model (used IITB as an additional resource for training)
12	CNLP-NITS-PP	MMCHTEXT24en-hi	2022/07/11 12:43:47	6742	-	-	-	-	-	-	0.770640	-	-	-	NMT	No	Transliteration-based phrase pairs augmentation in training using BRNN-based NMT
13	CNLP-NITS-PP	MMCHTEXT24en-hi	2021/04/28 00:15:03	5732	-	-	-	-	-	-	0.770621	-	-	-	NMT	Yes	Pretrained monolingual data (IITB) and fine-tuned with parallel data (WAT21 train data+ Extracted Phrase pairs from WAT21 train data +IITB train data) in training using BRNN model.
14	DCU_NMT	MMCHTEXT24en-hi	2024/08/11 22:57:15	7349	-	-	-	-	-	-	0.762839	-	-	-	NMT	No	En-Hi system trained using backtranslated data from flicker.
15	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:12:48	4188	-	-	-	-	-	-	0.736262	-	-	-	NMT	No
16	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:15:08	4189	-	-	-	-	-	-	0.736262	-	-	-	NMT	No
17	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:18:17	4190	-	-	-	-	-	-	0.736262	-	-	-	NMT	No
18	SILO_NLP	MMCHTEXT24en-hi	2022/07/12 04:15:44	6838	-	-	-	-	-	-	0.728801	-	-	-	NMT	No	Fine-tuning with pre-trained mBART-50 model
19	CNLP-NITS	MMCHTEXT24en-hi	2020/09/18 16:12:32	3898	-	-	-	-	-	-	0.714980	-	-	-	NMT	Yes	Pretrained monolingual data (IITB) and fine-tuned with parallel data in training using BRNN model.
20	DCU_NMT	MMCHTEXT24en-hi	2024/08/11 23:00:10	7350	-	-	-	-	-	-	0.709336	-	-	-	NMT	No	The baseline text-only model trained only on provided data.
21	ORGANIZER	MMCHTEXT24en-hi	2020/08/27 21:08:36	3587	-	-	-	-	-	-	0.691315	-	-	-	NMT	No	Transformer base model
22		MMCHTEXT24en-hi	2020/09/19 06:17:42	4031	-	-	-	-	-	-	0.623644	-	-	-	NMT	No	For the text-only Eng-Hindi translation task, we use an adaptation of the NMT-Keras code to suit our problem. Here, the focus is on long term translation as well as active learning strategies. The t
23	ODIAGEN	MMCHTEXT24en-hi	2024/08/15 17:16:06	7402	-	-	-	-	-	-	0.151195	-	-	-	Other	No	LLM based 8fine-tuned multimodal Llava model for region-specific instruction set in Hindi

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	Volta	MMCHTEXT24en-hi	2021/05/25 13:55:36	6429	-	-	-	-	-	-	0.876300	-	-	-	NMT	Yes	Finetuned mBART (Used IITB corpus for data augmentation)
2	ODIANLP	MMCHTEXT24en-hi	2020/09/14 19:07:05	3713	-	-	-	-	-	-	0.804880	-	-	-	NMT	Yes	Transformer model (used IITB as an additional resource for training)
3	CNLP-NITS-PP	MMCHTEXT24en-hi	2021/04/28 00:15:03	5732	-	-	-	-	-	-	0.797409	-	-	-	NMT	Yes	Pretrained monolingual data (IITB) and fine-tuned with parallel data (WAT21 train data+ Extracted Phrase pairs from WAT21 train data +IITB train data) in training using BRNN model.
4	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:12:48	4188	-	-	-	-	-	-	0.773230	-	-	-	NMT	No
5	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:15:08	4189	-	-	-	-	-	-	0.773230	-	-	-	NMT	No
6	ORGANIZER	MMCHTEXT24en-hi	2020/11/07 02:18:17	4190	-	-	-	-	-	-	0.773230	-	-	-	NMT	No
7	CNLP-NITS	MMCHTEXT24en-hi	2020/09/18 16:12:32	3898	-	-	-	-	-	-	0.750320	-	-	-	NMT	Yes	Pretrained monolingual data (IITB) and fine-tuned with parallel data in training using BRNN model.
8	ORGANIZER	MMCHTEXT24en-hi	2020/08/27 21:08:36	3587	-	-	-	-	-	-	0.727010	-	-	-	NMT	No	Transformer base model
9		MMCHTEXT24en-hi	2020/09/19 06:17:42	4031	-	-	-	-	-	-	0.698600	-	-	-	NMT	No	For the text-only Eng-Hindi translation task, we use an adaptation of the NMT-Keras code to suit our problem. Here, the focus is on long term translation as well as active learning strategies. The t
10	nlp_novices	MMCHTEXT24en-hi	2022/07/10 22:51:34	6725	-	-	-	-	-	-	0.000000	-	-	-	NMT	Yes	Finetuned Transformers over OPUS Corpora additionally
11	CNLP-NITS-PP	MMCHTEXT24en-hi	2022/07/11 12:43:47	6742	-	-	-	-	-	-	0.000000	-	-	-	NMT	No	Transliteration-based phrase pairs augmentation in training using BRNN-based NMT
12	SILO_NLP	MMCHTEXT24en-hi	2022/07/12 04:15:44	6838	-	-	-	-	-	-	0.000000	-	-	-	NMT	No	Fine-tuning with pre-trained mBART-50 model
13	ODIAGEN	MMCHTEXT24en-hi	2023/07/03 02:56:51	7088	-	-	-	-	-	-	0.000000	-	-	-	NMT	No	Fine-tuning Transformer using NLLB-200 from Facebook
14	ODIAGEN	MMCHTEXT24en-hi	2023/07/06 12:31:36	7110	-	-	-	-	-	-	0.000000	-	-	-	NMT	No	Fine-tuning Transformer using NLLB-200 from Facebook
15	00-7	MMCHTEXT24en-hi	2024/08/09 18:03:29	7313	-	-	-	-	-	-	0.000000	-	-	-	NMT	No
16	DCU_NMT	MMCHTEXT24en-hi	2024/08/11 22:57:15	7349	-	-	-	-	-	-	0.000000	-	-	-	NMT	No	En-Hi system trained using backtranslated data from flicker.
17	DCU_NMT	MMCHTEXT24en-hi	2024/08/11 23:00:10	7350	-	-	-	-	-	-	0.000000	-	-	-	NMT	No	The baseline text-only model trained only on provided data.
18	ODIAGEN	MMCHTEXT24en-hi	2024/08/12 23:38:25	7358	-	-	-	-	-	-	0.000000	-	-	-	Other	No	LLM based (fine-tuning Mistral7B)
19	ODIAGEN	MMCHTEXT24en-hi	2024/08/15 17:16:06	7402	-	-	-	-	-	-	0.000000	-	-	-	Other	No	LLM based 8fine-tuned multimodal Llava model for region-specific instruction set in Hindi
20	IITP-AI-NLP-ML	MMCHTEXT24en-hi	2025/10/22 22:13:17	7471	-	-	-	-	-	-	0.000000	-	-	-	NMT	Yes	Finetuned indicTrans on all 4 languages combined
21	OdiaGenAI	MMCHTEXT24en-hi	2025/10/23 11:39:05	7485	-	-	-	-	-	-	0.000000	-	-	-		Yes	NLLB-3.3B fine-tuned with WAT dataset + 100k pairs from samanantar for Hindi language only.
22	BLEU Monday	MMCHTEXT24en-hi	2025/10/23 16:16:00	7490	-	-	-	-	-	-	0.000000	-	-	-	Other	Yes	LORA on corrected data
23	BLEU Monday	MMCHTEXT24en-hi	2025/10/23 19:25:18	7500	-	-	-	-	-	-	0.000000	-	-	-	Other	Yes	LORA on original data

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