opus2m-2020-08-02.yml

release: eng-zle/opus2m-2020-08-02.zip
release-date: 2020-08-02
dataset-name: opus2m
modeltype: transformer
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   - source: spm32k
   - target: spm32k
subword-models:
   - source: source.spm
   - target: target.spm
source-languages:
   - eng
target-languages:
   - bel
   - orv
   - rue
   - rus
   - ukr
use-target-labels:
   - >>bel<<
   - >>bel_Latn<<
   - >>orv_Cyrl<<
   - >>rue<<
   - >>rus<<
   - >>ukr<<
test-data:
   - Tatoeba-test.eng-bel: 2500/12908
   - Tatoeba-test.eng-multi: 10000/50373
   - Tatoeba-test.eng-orv: 322/1364
   - Tatoeba-test.eng-rue: 120/369
   - Tatoeba-test.eng-rus: 10000/53367
   - Tatoeba-test.eng-ukr: 10000/47921
   - newstest2012.eng-rus: 3003/53938
   - newstest2013.eng-rus: 3000/48639
   - newstest2015-enru.eng-rus: 2818/46352
   - newstest2016-enru.eng-rus: 2998/51923
   - newstest2017-enru.eng-rus: 3001/50385
   - newstest2018-enru.eng-rus: 3000/51988
   - newstest2019-enru.eng-rus: 1997/40379
BLEU-scores:
   - Tatoeba-test.eng-bel: 20.8
   - Tatoeba-test.eng-multi: 37.2
   - Tatoeba-test.eng-orv: 0.6
   - Tatoeba-test.eng-rue: 1.4
   - Tatoeba-test.eng-rus: 41.3
   - Tatoeba-test.eng-ukr: 38.7
   - newstest2012.eng-rus: 27.4
   - newstest2013.eng-rus: 21.4
   - newstest2015-enru.eng-rus: 24.2
   - newstest2016-enru.eng-rus: 23.3
   - newstest2017-enru.eng-rus: 25.3
   - newstest2018-enru.eng-rus: 22.4
   - newstest2019-enru.eng-rus: 24.1
chr-F-scores:
   - Tatoeba-test.eng-bel: 0.471
   - Tatoeba-test.eng-multi: 0.580
   - Tatoeba-test.eng-orv: 0.130
   - Tatoeba-test.eng-rue: 0.168
   - Tatoeba-test.eng-rus: 0.616
   - Tatoeba-test.eng-ukr: 0.596
   - newstest2012.eng-rus: 0.550
   - newstest2013.eng-rus: 0.493
   - newstest2015-enru.eng-rus: 0.534
   - newstest2016-enru.eng-rus: 0.518
   - newstest2017-enru.eng-rus: 0.541
   - newstest2018-enru.eng-rus: 0.527
   - newstest2019-enru.eng-rus: 0.505