opus1m+bt-2021-05-01.yml

release: dra-eng/opus1m+bt-2021-05-01.zip
release-date: 2021-05-01
dataset-name: opus1m+bt
modeltype: transformer-align
vocabulary:
   source: opus1m+bt.spm32k-spm32k.vocab.yml
   target: opus1m+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:
   - kan
   - mal
   - tam
   - tel
target-languages:
   - eng
training-data:
   kan-eng: Tatoeba-train (404471) 
   mal-eng: Tatoeba-train (1000000) wikibooks.aa (986780) wikinews.aa (455200) wikipedia.aa (972748) wikipedia.ab (972828) wikipedia.ac (972902) wikipedia.ad (972646) wikiquote.aa (994815) 
   tam-eng: Tatoeba-train (1000000) wikibooks.aa (986780) wikinews.aa (455200) wikipedia.aa (972748) wikipedia.ab (972828) wikipedia.ac (972902) wikipedia.ad (972646) wikiquote.aa (994815) 
   tam_Deva-eng: Tatoeba-train (4) 
   tel-eng: Tatoeba-train (328547) 
validation-data:
   eng-kan: Tatoeba-dev, 993
   eng-mal: Tatoeba-dev, 990
   eng-tam: Tatoeba-dev, 996
   eng-tel: Tatoeba-dev, 992
   total-size-shuffled: 3963
   devset-selected: top 3963  lines of Tatoeba-dev.src.shuffled
test-data:
   Tatoeba-test.kan-eng: 167/1252
   Tatoeba-test.mal-eng: 802/5558
   Tatoeba-test.multi-eng: 1541/10641
   Tatoeba-test.tam-eng: 311/2125
   Tatoeba-test.tel-eng: 261/1706
   tico19-test.tam-eng: 2100/56848
BLEU-scores:
   Tatoeba-test.kan-eng: 17.8
   Tatoeba-test.mal-eng: 42.8
   Tatoeba-test.multi-eng: 32.2
   Tatoeba-test.tam-eng: 27.4
   Tatoeba-test.tel-eng: 19.5
   tico19-test.tam-eng: 13.6
chr-F-scores:
   Tatoeba-test.kan-eng: 0.404
   Tatoeba-test.mal-eng: 0.602
   Tatoeba-test.multi-eng: 0.519
   Tatoeba-test.tam-eng: 0.460
   Tatoeba-test.tel-eng: 0.405
   tico19-test.tam-eng: 0.380