opus+bt-2021-04-30.yml

release: fra-eng/opus+bt-2021-04-30.zip
release-date: 2021-04-30
dataset-name: opus+bt
modeltype: transformer-align
vocabulary:
   source: opus+bt.spm32k-spm32k.vocab.yml
   target: opus+bt.spm32k-spm32k.vocab.yml
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   source: spm32k
   target: spm32k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - fra
target-languages:
   - eng
training-data:
   fra-eng: Tatoeba-train (180923857) wikinews.aa (457553) wikipedia.aa (984195) wikipedia.ab (984396) wikipedia.ac (984310) wikipedia.ad (984106) wikiquote.aa (997127) 
validation-data:
   eng-fra: Tatoeba-dev, 250098
   total-size-shuffled: 249757
   devset-selected: top 5000  lines of Tatoeba-dev.src.shuffled!
test-data:
   newsdiscussdev2015-enfr.fra-eng: 1500/27759
   newsdiscusstest2015-enfr.fra-eng: 1500/26995
   newssyscomb2009.fra-eng: 502/11821
   news-test2008.fra-eng: 2051/49380
   newstest2009.fra-eng: 2525/65402
   newstest2010.fra-eng: 2489/61724
   newstest2011.fra-eng: 3003/74681
   newstest2012.fra-eng: 3003/72812
   newstest2013.fra-eng: 3000/64505
   newstest2014-fren.fra-eng: 3003/70708
   Tatoeba-test.fra-eng: 10000/77174
   tico19-test.fra-eng: 2100/56347
BLEU-scores:
   newsdiscussdev2015-enfr.fra-eng: 33.9
   newsdiscusstest2015-enfr.fra-eng: 38.8
   newssyscomb2009.fra-eng: 30.6
   news-test2008.fra-eng: 26.2
   newstest2009.fra-eng: 30.2
   newstest2010.fra-eng: 32.5
   newstest2011.fra-eng: 33.0
   newstest2012.fra-eng: 32.9
   newstest2013.fra-eng: 33.9
   newstest2014-fren.fra-eng: 38.2
   Tatoeba-test.fra-eng: 57.6
   tico19-test.fra-eng: 40.2
chr-F-scores:
   newsdiscussdev2015-enfr.fra-eng: 0.584
   newsdiscusstest2015-enfr.fra-eng: 0.618
   newssyscomb2009.fra-eng: 0.572
   news-test2008.fra-eng: 0.542
   newstest2009.fra-eng: 0.569
   newstest2010.fra-eng: 0.591
   newstest2011.fra-eng: 0.597
   newstest2012.fra-eng: 0.591
   newstest2013.fra-eng: 0.591
   newstest2014-fren.fra-eng: 0.635
   Tatoeba-test.fra-eng: 0.721
   tico19-test.fra-eng: 0.633