opus-2021-02-19.yml

release: eng-fin/opus-2021-02-19.zip
release-date: 2021-02-19
dataset-name: opus
modeltype: transformer-align
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   source: spm32k
   target: spm32k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - eng
target-languages:
   - fin
training-data:
   eng-fin: Tatoeba-train (45205537) 
validation-data:
   eng-fin: Tatoeba-dev, 69721
   total-size-shuffled: 69721
   devset-selected: top 5000  lines of Tatoeba-dev.src.shuffled!
test-data:
   newsdev2015-enfi.eng-fin: 1500/23375
   newstest2015-enfi.eng-fin: 1370/19968
   newstest2016-enfi.eng-fin: 3000/48116
   newstest2017-enfi.eng-fin: 3002/45718
   newstest2018-enfi.eng-fin: 3000/45475
   newstest2019-enfi.eng-fin: 1997/38369
   newstestB2016-enfi.eng-fin: 3000/45766
   newstestB2017-enfi.eng-fin: 3002/45506
   Tatoeba-test.eng-fin: 10000/60517
BLEU-scores:
   newsdev2015-enfi.eng-fin: 21.6
   newstest2015-enfi.eng-fin: 23.2
   newstest2016-enfi.eng-fin: 24.9
   newstest2017-enfi.eng-fin: 27.5
   newstest2018-enfi.eng-fin: 18.4
   newstest2019-enfi.eng-fin: 23.3
   newstestB2016-enfi.eng-fin: 19.7
   newstestB2017-enfi.eng-fin: 22.7
   Tatoeba-test.eng-fin: 38.7
chr-F-scores:
   newsdev2015-enfi.eng-fin: 0.556
   newstest2015-enfi.eng-fin: 0.567
   newstest2016-enfi.eng-fin: 0.578
   newstest2017-enfi.eng-fin: 0.605
   newstest2018-enfi.eng-fin: 0.532
   newstest2019-enfi.eng-fin: 0.551
   newstestB2016-enfi.eng-fin: 0.542
   newstestB2017-enfi.eng-fin: 0.565
   Tatoeba-test.eng-fin: 0.629