opus-2021-02-24.yml

release: ita-msa/opus-2021-02-24.zip
release-date: 2021-02-24
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<<"
   - ">>msa_Latn<<"
   - ">>zlm<<"
   - ">>zlm_Latn<<"
   - ">>zsm_Latn<<"
source-languages:
   - ita
target-languages:
   - ind
   - jak
   - msa
   - zlm
   - zsm
training-data:
   ita-ind: Tatoeba-train (6076279) 
   ita-jak_Latn: Tatoeba-train (6536) 
   ita-msa_Latn: Tatoeba-train (1414818) 
   ita-zlm: Tatoeba-train (171) 
   ita-zlm_Latn: Tatoeba-train (73963) 
validation-data:
   ind-ita: Tatoeba-dev, 817
   ita-jak_Latn: Tatoeba-dev, 2
   ita-msa_Latn: Tatoeba-dev, 172
   ita-zlm_Latn: Tatoeba-dev, 9
   total-size-shuffled: 995
   devset-selected: top 995  lines of Tatoeba-dev.src.shuffled!
test-data:
   Tatoeba-test.ita-ind: 368/2333
   Tatoeba-test.ita-zsm_Latn: 58/432
   Tatoeba-test.ita-msa: 426/2765
BLEU-scores:
   Tatoeba-test.ita-ind: 28.7
   Tatoeba-test.ita-zsm_Latn: 22.2
   Tatoeba-test.ita-msa: 28.0
chr-F-scores:
   Tatoeba-test.ita-ind: 0.563
   Tatoeba-test.ita-zsm_Latn: 0.479
   Tatoeba-test.ita-msa: 0.551