NICT_LOGO.JPG KYOTO-U_LOGO.JPG

WAT

The Workshop on Asian Translation
Evaluation Results

[EVALUATION RESULTS TOP] | [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]

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
1BITS-PMMEVMM24en-hi2023/07/08 13:46:537125------45.00---NMTYesNLLB model finetuned on captions + object tags of original & synthetic images using DETR model
2IITP-AI-NLP-MLMMEVMM24en-hi2025/10/22 21:31:417456------44.90---SMTYesUsed Selective Attention Architecture with IndicTrans as the base model and CLIP ViT-B/16 model to extract image features. We extract a) Full image feats, and b) cropped image feats and pick the one w
3VoltaMMEVMM24en-hi2021/05/25 13:55:526428------44.64---NMTYesFinetuned mBART (Used IITB for data augmentation) and added object tags to the input using Mask RCNN
4v036MMEVMM24en-hi2024/08/15 19:27:237411------44.60---NMTNo
500-7MMEVMM24en-hi2024/08/11 13:50:497325------43.70---NMTNo
6v036MMEVMM24en-hi2024/08/15 18:52:027408------42.90---NMTNo
7iitpMMEVMM24en-hi2021/05/01 23:19:145941------42.47---NMTNoRemoved special chars at start and end of sentnce 1. pre-trained with HindEnCorp, trained with Vis Gen 2. trained with Visual Genome Selected best of two for each sentence according to translation
8SILO_NLPMMEVMM24en-hi2022/07/18 17:31:416958------42.00---NMTYesObject Tags (Image) + Flickr8 dataset as additional resource + Finetune mBART
9v036MMEVMM24en-hi2024/08/11 21:39:077345------41.80---NMTNoNMT based system using both image descriptors and text description. A multistage LLM pipeline used for extracting image data descriptions and translation. Fine tuning done in few cases Models Used:
10ODIAGENMMEVMM24en-hi2023/07/06 03:45:527105------41.60---NMTNoImage features extracted as Object tags appended with text and MBART fine-tuning
11DCU_NMTMMEVMM24en-hi2024/08/11 23:03:337351------40.60---NMTNoContext-aware model that uses image caption data extracted from images as context.
12DCU_NMTMMEVMM24en-hi2024/08/13 03:24:097371------40.60---NMTNoNMT system trained on constrained resources using bert encoded context extracted from visual representation of training data. The context is used only on source side.
13CNLP-NITSMMEVMM24en-hi2020/09/18 15:59:023896------40.51---NMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.
14UNLPMMEVMM24en-hi2024/08/13 17:57:397392------40.30---NMTNoUsing the Transformer-based Gated Fusion model to integrate both text and visual data.
15CNLP-NITS-PPMMEVMM24en-hi2021/04/28 00:05:195731------39.46---NMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data (WAT21 train data+ Extracted Phrase pairs from WAT21 train data +IITB train data) and visual features in training using
16CNLP-NITS-PPMMEVMM24en-hi2022/07/11 12:31:326740------39.40---NMTNoTransliteration-based phrase pairs augmentation and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.
17v036MMEVMM24en-hi2024/08/14 20:34:407398------39.00---NMTNo
18ORGANIZERMMEVMM24en-hi2020/11/07 01:54:464183------38.63---NMTNo
19239233MMEVMM24en-hi2024/08/13 07:30:137379------29.70---NMTYesOne-shot prompt for synthetic QA description from captions; translate QA using IndicTrans2; generate caption from QA as context
20ORGANIZERMMEVMM24en-hi2020/11/07 01:49:404182------27.41---NMTNo

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
1v036MMEVMM24en-hi2024/08/15 19:27:237411------0.833853---NMTNo
2IITP-AI-NLP-MLMMEVMM24en-hi2025/10/22 21:31:417456------0.830773---SMTYesUsed Selective Attention Architecture with IndicTrans as the base model and CLIP ViT-B/16 model to extract image features. We extract a) Full image feats, and b) cropped image feats and pick the one w
3BITS-PMMEVMM24en-hi2023/07/08 13:46:537125------0.829320---NMTYesNLLB model finetuned on captions + object tags of original & synthetic images using DETR model
4v036MMEVMM24en-hi2024/08/15 18:52:027408------0.828421---NMTNo
5VoltaMMEVMM24en-hi2021/05/25 13:55:526428------0.823319---NMTYesFinetuned mBART (Used IITB for data augmentation) and added object tags to the input using Mask RCNN
600-7MMEVMM24en-hi2024/08/11 13:50:497325------0.813357---NMTNo
7ODIAGENMMEVMM24en-hi2023/07/06 03:45:527105------0.811420---NMTNoImage features extracted as Object tags appended with text and MBART fine-tuning
8v036MMEVMM24en-hi2024/08/11 21:39:077345------0.809613---NMTNoNMT based system using both image descriptors and text description. A multistage LLM pipeline used for extracting image data descriptions and translation. Fine tuning done in few cases Models Used:
9iitpMMEVMM24en-hi2021/05/01 23:19:145941------0.807123---NMTNoRemoved special chars at start and end of sentnce 1. pre-trained with HindEnCorp, trained with Vis Gen 2. trained with Visual Genome Selected best of two for each sentence according to translation
10DCU_NMTMMEVMM24en-hi2024/08/11 23:03:337351------0.806358---NMTNoContext-aware model that uses image caption data extracted from images as context.
11CNLP-NITSMMEVMM24en-hi2020/09/18 15:59:023896------0.803208---NMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.
12CNLP-NITS-PPMMEVMM24en-hi2022/07/11 12:31:326740------0.802635---NMTNoTransliteration-based phrase pairs augmentation and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.
13CNLP-NITS-PPMMEVMM24en-hi2021/04/28 00:05:195731------0.802055---NMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data (WAT21 train data+ Extracted Phrase pairs from WAT21 train data +IITB train data) and visual features in training using
14UNLPMMEVMM24en-hi2024/08/13 17:57:397392------0.800532---NMTNoUsing the Transformer-based Gated Fusion model to integrate both text and visual data.
15DCU_NMTMMEVMM24en-hi2024/08/13 03:24:097371------0.799818---NMTNoNMT system trained on constrained resources using bert encoded context extracted from visual representation of training data. The context is used only on source side.
16SILO_NLPMMEVMM24en-hi2022/07/18 17:31:416958------0.796441---NMTYesObject Tags (Image) + Flickr8 dataset as additional resource + Finetune mBART
17v036MMEVMM24en-hi2024/08/14 20:34:407398------0.790548---NMTNo
18ORGANIZERMMEVMM24en-hi2020/11/07 01:54:464183------0.767422---NMTNo
19239233MMEVMM24en-hi2024/08/13 07:30:137379------0.725450---NMTYesOne-shot prompt for synthetic QA description from captions; translate QA using IndicTrans2; generate caption from QA as context
20ORGANIZERMMEVMM24en-hi2020/11/07 01:49:404182------0.651455---NMTNo

Notice:

Back to top

AMFM


# Team Task Date/Time DataID AMFM
Method
 Other
Resources
 System
Description
unuse unuse unuse unuse unuse unuse unuse unuse unuse unuse
1VoltaMMEVMM24en-hi2021/05/25 13:55:526428------0.839100---NMTYesFinetuned mBART (Used IITB for data augmentation) and added object tags to the input using Mask RCNN
2CNLP-NITSMMEVMM24en-hi2020/09/18 15:59:023896------0.820980---NMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.
3ORGANIZERMMEVMM24en-hi2020/11/07 01:54:464183------0.772870---NMTNo
4ORGANIZERMMEVMM24en-hi2020/11/07 01:49:404182------0.691950---NMTNo
5CNLP-NITS-PPMMEVMM24en-hi2021/04/28 00:05:195731------0.641430---NMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data (WAT21 train data+ Extracted Phrase pairs from WAT21 train data +IITB train data) and visual features in training using
6iitpMMEVMM24en-hi2021/05/01 23:19:145941------0.629444---NMTNoRemoved special chars at start and end of sentnce 1. pre-trained with HindEnCorp, trained with Vis Gen 2. trained with Visual Genome Selected best of two for each sentence according to translation
7CNLP-NITS-PPMMEVMM24en-hi2022/07/11 12:31:326740------0.000000---NMTNoTransliteration-based phrase pairs augmentation and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.
8SILO_NLPMMEVMM24en-hi2022/07/18 17:31:416958------0.000000---NMTYesObject Tags (Image) + Flickr8 dataset as additional resource + Finetune mBART
9ODIAGENMMEVMM24en-hi2023/07/06 03:45:527105------0.000000---NMTNoImage features extracted as Object tags appended with text and MBART fine-tuning
10BITS-PMMEVMM24en-hi2023/07/08 13:46:537125------0.000000---NMTYesNLLB model finetuned on captions + object tags of original & synthetic images using DETR model
1100-7MMEVMM24en-hi2024/08/11 13:50:497325------0.000000---NMTNo
12v036MMEVMM24en-hi2024/08/11 21:39:077345------0.000000---NMTNoNMT based system using both image descriptors and text description. A multistage LLM pipeline used for extracting image data descriptions and translation. Fine tuning done in few cases Models Used:
13DCU_NMTMMEVMM24en-hi2024/08/11 23:03:337351------0.000000---NMTNoContext-aware model that uses image caption data extracted from images as context.
14DCU_NMTMMEVMM24en-hi2024/08/13 03:24:097371------0.000000---NMTNoNMT system trained on constrained resources using bert encoded context extracted from visual representation of training data. The context is used only on source side.
15239233MMEVMM24en-hi2024/08/13 07:30:137379------0.000000---NMTYesOne-shot prompt for synthetic QA description from captions; translate QA using IndicTrans2; generate caption from QA as context
16UNLPMMEVMM24en-hi2024/08/13 17:57:397392------0.000000---NMTNoUsing the Transformer-based Gated Fusion model to integrate both text and visual data.
17v036MMEVMM24en-hi2024/08/14 20:34:407398------0.000000---NMTNo
18v036MMEVMM24en-hi2024/08/15 18:52:027408------0.000000---NMTNo
19v036MMEVMM24en-hi2024/08/15 19:27:237411------0.000000---NMTNo
20IITP-AI-NLP-MLMMEVMM24en-hi2025/10/22 21:31:417456------0.000000---SMTYesUsed Selective Attention Architecture with IndicTrans as the base model and CLIP ViT-B/16 model to extract image features. We extract a) Full image feats, and b) cropped image feats and pick the one w

Notice:

Back to top

HUMAN (WAT2022)


# Team Task Date/Time DataID HUMAN
Method
 Other
Resources
System
Description
1CNLP-NITS-PPMMEVMM24en-hi2022/07/11 12:31:326740UnderwayNMTNoTransliteration-based phrase pairs augmentation and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.
2SILO_NLPMMEVMM24en-hi2022/07/18 17:31:416958UnderwayNMTYesObject Tags (Image) + Flickr8 dataset as additional resource + Finetune mBART

Notice:
Back to top

HUMAN (WAT2021)


# Team Task Date/Time DataID HUMAN
Method
 Other
Resources
System
Description
1CNLP-NITS-PPMMEVMM24en-hi2021/04/28 00:05:195731UnderwayNMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data (WAT21 train data+ Extracted Phrase pairs from WAT21 train data +IITB train data) and visual features in training using
2iitpMMEVMM24en-hi2021/05/01 23:19:145941UnderwayNMTNoRemoved special chars at start and end of sentnce 1. pre-trained with HindEnCorp, trained with Vis Gen 2. trained with Visual Genome Selected best of two for each sentence according to translation

Notice:
Back to top

HUMAN (WAT2020)


# Team Task Date/Time DataID HUMAN
Method
 Other
Resources
System
Description
1CNLP-NITSMMEVMM24en-hi2020/09/18 15:59:023896UnderwayNMTYesPretrained monolingual data (IITB) using Glove and fine-tuned with parallel data and visual features in training using BRNN encoder and doubly-attentive-rnn decoder.

Notice:
Back to top

HUMAN (WAT2019)


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

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

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