opus-2021-02-18.yml

release: deu-msa/opus-2021-02-18.zip
release-date: 2021-02-18
dataset-name: opus
modeltype: transformer
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   source: spm32k
   target: spm32k
subword-models:
   source: source.spm
   target: target.spm
use-target-labels:
   - ">>ind<<"
   - ">>jak_Latn<<"
   - ">>min<<"
   - ">>msa_Latn<<"
   - ">>zlm<<"
   - ">>zlm_Latn<<"
   - ">>zsm_Latn<<"
source-languages:
   - deu
target-languages:
   - ind
   - jak
   - min
   - msa
   - zlm
   - zsm
training-data:
   deu-ind: Tatoeba-train (5319357) 
   deu-jak_Latn: Tatoeba-train (6601) 
   deu-min: Tatoeba-train (8) 
   deu-msa_Latn: Tatoeba-train (1402955) 
   deu-zlm: Tatoeba-train (342) 
   deu-zlm_Latn: Tatoeba-train (77905) 
validation-data:
   deu-ind: Tatoeba-dev, 787
   deu-jak_Latn: Tatoeba-dev, 1
   deu-msa_Latn: Tatoeba-dev, 202
   deu-zlm_Latn: Tatoeba-dev, 10
   total-size-shuffled: 995
   devset-selected: top 995  lines of Tatoeba-dev.src.shuffled!
test-data:
   Tatoeba-test.deu-ind: 488/3359
   Tatoeba-test.deu-zsm_Latn: 47/370
   Tatoeba-test.deu-msa: 535/3729
BLEU-scores:
   Tatoeba-test.deu-ind: 36.6
   Tatoeba-test.deu-zsm_Latn: 16.1
   Tatoeba-test.deu-msa: 34.8
chr-F-scores:
   Tatoeba-test.deu-ind: 0.631
   Tatoeba-test.deu-zsm_Latn: 0.461
   Tatoeba-test.deu-msa: 0.613