WMT2017 German-English machine translation challenge for news

Translate news articles from German into English. [ver. 1.0.0]

Git repo URL: git://gonito.net/wmt-2017 / Branch: master
Run git clone --single-branch git://gonito.net/wmt-2017 -b master to get the challenge data
Browse at https://gonito.net/gitlist/wmt-2017.git/master

Leaderboard

# submitter when ver. description test-A BLEU ×
1 [anonymized] 2020-01-15 09:58 1.0.0 Poprawienie Tokenizacji istniejacego rozwiazania v3 ready-made fairseq 0.41504 3
2 [anonymized] 2020-01-14 20:53 1.0.0 fix tokenization of output ready-made fairseq 0.39879 1
3 [anonymized] 2021-02-17 19:16 1.0.0 result fairseq m2m-100 just-inference 0.35347 1
4 [anonymized] 2021-02-12 19:54 1.0.0 notebook fairseq m2m-100 just-inference 0.35347 2
5 p/tlen 2018-01-09 18:10 1.0.0 WMT16 neural model (decoded with Amun) + de-escape apostrophes neural-network 0.28988 7
6 [anonymized] 2020-01-07 11:35 1.0.0 ready-made Fairseq model fairseq ready-made-model 0.26579 1
7 [anonymized] 2019-12-30 09:59 1.0.0 Runed a ready-made Fairseq model fairseq ready-made-model 0.26579 1
8 [anonymized] 2018-02-15 00:38 1.0.0 Tensorflow 80k iterations ; beam 4 alpha 0.9 ready-made neural-network 0.26322 4
9 kaczla 2017-11-29 11:33 1.0.0 baseline Moses on 100% utterances + 40GB english monolingual data (4-gram + pruning + 9 iteration MERT - weights no change) 0.25513 26
10 [anonymized] 2019-05-22 12:01 1.0.0 marian 1M neural-network marian 0.24561 41
11 [anonymized] 2020-01-28 17:56 1.0.0 translation with ready made fairseq transformer.wmt19.de-en v2 fairseq ready-made-model 0.23698 2
12 [anonymized] 2017-11-22 15:05 1.0.0 40GB language model ready-made moses 0.23152 5
13 [anonymized] 2018-01-03 00:06 1.0.0 dev-0 dev-1 test-A -stack 1500 moses 0.20230 14
14 [anonymized] 2018-01-31 11:43 1.0.0 corpus=590616, NB_OF_EPOCHS=8, MAX_WORDS=46000 neural-network 0.19461 14
15 [anonymized] 2018-02-07 11:10 1.0.0 Add 5G data moses 0.19183 5
16 [anonymized] 2017-11-26 09:13 1.0.0 5G monolingual data moses 0.18459 16
17 [anonymized] 2018-01-08 16:35 1.0.0 TAU-2017-21 - improve solution by changing some decoding options: -stack 150 moses 0.18379 14
18 [anonymized] 2020-01-27 12:19 1.0.0 CNN, sample_size = 5mln, epochs = 5 fairseq train 0.07834 21
19 [anonymized] 2018-02-04 23:59 1.0.0 'baseline' moses 0.02823 5
20 [anonymized] 2017-10-03 13:23 1.0.0 just checkin' stupid 0.00762 1