opus2m-2020-08-01.yml

release: eng-phi/opus2m-2020-08-01.zip
release-date: 2020-08-01
dataset-name: opus2m
modeltype: transformer
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   source: spm32k
   target: spm32k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - eng
target-languages:
   - akl
   - ceb
   - hil
   - ilo
   - pag
   - war
use-target-labels:
   - ">>akl_Latn<<"
   - ">>ceb<<"
   - ">>hil<<"
   - ">>ilo<<"
   - ">>pag<<"
   - ">>war<<"
test-data:
   Tatoeba-test.eng-akl: 27/69
   Tatoeba-test.eng-ceb: 378/1678
   Tatoeba-test.eng-hil: 22/104
   Tatoeba-test.eng-ilo: 1093/6040
   Tatoeba-test.eng-multi: 4578/25022
   Tatoeba-test.eng-pag: 49/268
   Tatoeba-test.eng-war: 1512/9364
BLEU-scores:
   Tatoeba-test.eng-akl: 7.1
   Tatoeba-test.eng-ceb: 10.5
   Tatoeba-test.eng-hil: 18.0
   Tatoeba-test.eng-ilo: 33.4
   Tatoeba-test.eng-multi: 13.1
   Tatoeba-test.eng-pag: 19.4
   Tatoeba-test.eng-war: 12.8
chr-F-scores:
   Tatoeba-test.eng-akl: 0.245
   Tatoeba-test.eng-ceb: 0.435
   Tatoeba-test.eng-hil: 0.506
   Tatoeba-test.eng-ilo: 0.590
   Tatoeba-test.eng-multi: 0.392
   Tatoeba-test.eng-pag: 0.481
   Tatoeba-test.eng-war: 0.441