opus2m-2020-08-01.yml

release: eng-sla/opus2m-2020-08-01.zip
release-date: 2020-08-01
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
   - bos
   - bul
   - ces
   - csb
   - dsb
   - hrv
   - hsb
   - mkd
   - orv
   - pol
   - rue
   - rus
   - slv
   - srp
   - ukr
use-target-labels:
   - ">>bel<<"
   - ">>bel_Latn<<"
   - ">>bos_Latn<<"
   - ">>bul<<"
   - ">>bul_Latn<<"
   - ">>ces<<"
   - ">>csb_Latn<<"
   - ">>dsb<<"
   - ">>hrv<<"
   - ">>hsb<<"
   - ">>mkd<<"
   - ">>orv_Cyrl<<"
   - ">>pol<<"
   - ">>rue<<"
   - ">>rus<<"
   - ">>slv<<"
   - ">>srp_Cyrl<<"
   - ">>srp_Latn<<"
   - ">>ukr<<"
test-data:
   Tatoeba-test.eng-bel: 2500/12908
   Tatoeba-test.eng-bul: 10000/56740
   Tatoeba-test.eng-ces: 10000/52129
   Tatoeba-test.eng-csb: 27/196
   Tatoeba-test.eng-dsb: 34/140
   Tatoeba-test.eng-hbs: 10000/52571
   Tatoeba-test.eng-hsb: 40/156
   Tatoeba-test.eng-mkd: 10000/51245
   Tatoeba-test.eng-multi: 10000/52312
   Tatoeba-test.eng-orv: 322/1364
   Tatoeba-test.eng-pol: 10000/52166
   Tatoeba-test.eng-rue: 120/369
   Tatoeba-test.eng-rus: 10000/53367
   Tatoeba-test.eng-slv: 2007/9410
   Tatoeba-test.eng-ukr: 10000/47921
   news-test2008.eng-ces: 2051/36403
   newssyscomb2009.eng-ces: 502/8592
   newstest2009.eng-ces: 2525/47071
   newstest2010.eng-ces: 2489/45422
   newstest2011.eng-ces: 3003/55301
   newstest2012.eng-ces: 3003/54732
   newstest2012.eng-rus: 3003/53938
   newstest2013.eng-ces: 3000/48472
   newstest2013.eng-rus: 3000/48639
   newstest2015-encs.eng-ces: 2656/39201
   newstest2015-enru.eng-rus: 2818/46352
   newstest2016-encs.eng-ces: 2999/48915
   newstest2016-enru.eng-rus: 2998/51923
   newstest2017-encs.eng-ces: 3005/46951
   newstest2017-enru.eng-rus: 3001/50385
   newstest2018-encs.eng-ces: 2983/47229
   newstest2018-enru.eng-rus: 3000/51988
   newstest2019-encs.eng-ces: 1997/37506
   newstest2019-enru.eng-rus: 1997/40379
BLEU-scores:
   Tatoeba-test.eng-bel: 22.9
   Tatoeba-test.eng-bul: 46.7
   Tatoeba-test.eng-ces: 42.7
   Tatoeba-test.eng-csb: 1.4
   Tatoeba-test.eng-dsb: 1.4
   Tatoeba-test.eng-hbs: 40.3
   Tatoeba-test.eng-hsb: 14.3
   Tatoeba-test.eng-mkd: 44.1
   Tatoeba-test.eng-multi: 41.0
   Tatoeba-test.eng-orv: 0.3
   Tatoeba-test.eng-pol: 42.0
   Tatoeba-test.eng-rue: 0.3
   Tatoeba-test.eng-rus: 40.5
   Tatoeba-test.eng-slv: 18.8
   Tatoeba-test.eng-ukr: 38.8
   news-test2008.eng-ces: 17.7
   newssyscomb2009.eng-ces: 20.1
   newstest2009.eng-ces: 19.1
   newstest2010.eng-ces: 19.3
   newstest2011.eng-ces: 20.4
   newstest2012.eng-ces: 18.3
   newstest2012.eng-rus: 27.4
   newstest2013.eng-ces: 21.5
   newstest2013.eng-rus: 20.9
   newstest2015-encs.eng-ces: 21.1
   newstest2015-enru.eng-rus: 24.5
   newstest2016-encs.eng-ces: 23.6
   newstest2016-enru.eng-rus: 23.0
   newstest2017-encs.eng-ces: 19.2
   newstest2017-enru.eng-rus: 25.0
   newstest2018-encs.eng-ces: 19.3
   newstest2018-enru.eng-rus: 22.3
   newstest2019-encs.eng-ces: 20.4
   newstest2019-enru.eng-rus: 24.0
chr-F-scores:
   Tatoeba-test.eng-bel: 0.489
   Tatoeba-test.eng-bul: 0.652
   Tatoeba-test.eng-ces: 0.624
   Tatoeba-test.eng-csb: 0.210
   Tatoeba-test.eng-dsb: 0.165
   Tatoeba-test.eng-hbs: 0.616
   Tatoeba-test.eng-hsb: 0.344
   Tatoeba-test.eng-mkd: 0.635
   Tatoeba-test.eng-multi: 0.610
   Tatoeba-test.eng-orv: 0.014
   Tatoeba-test.eng-pol: 0.637
   Tatoeba-test.eng-rue: 0.012
   Tatoeba-test.eng-rus: 0.612
   Tatoeba-test.eng-slv: 0.357
   Tatoeba-test.eng-ukr: 0.600
   news-test2008.eng-ces: 0.461
   newssyscomb2009.eng-ces: 0.484
   newstest2009.eng-ces: 0.479
   newstest2010.eng-ces: 0.483
   newstest2011.eng-ces: 0.486
   newstest2012.eng-ces: 0.461
   newstest2012.eng-rus: 0.551
   newstest2013.eng-ces: 0.489
   newstest2013.eng-rus: 0.490
   newstest2015-encs.eng-ces: 0.496
   newstest2015-enru.eng-rus: 0.536
   newstest2016-encs.eng-ces: 0.515
   newstest2016-enru.eng-rus: 0.519
   newstest2017-encs.eng-ces: 0.474
   newstest2017-enru.eng-rus: 0.541
   newstest2018-encs.eng-ces: 0.479
   newstest2018-enru.eng-rus: 0.526
   newstest2019-encs.eng-ces: 0.486
   newstest2019-enru.eng-rus: 0.506