opus2m-2020-08-12.yml

release: phi-eng/opus2m-2020-08-12.zip
release-date: 2020-08-12
dataset-name: opus2m
modeltype: transformer
pre-processing: normalization + SentencePiece (spm12k,spm12k)
subwords:
   source: spm12k
   target: spm12k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - akl
   - ceb
   - hil
   - ilo
   - pag
   - war
target-languages:
   - eng
test-data:
   Tatoeba-test.akl-eng: 27/60
   Tatoeba-test.ceb-eng: 378/1873
   Tatoeba-test.hil-eng: 22/127
   Tatoeba-test.ilo-eng: 1093/6184
   Tatoeba-test.multi-eng: 405/1933
   Tatoeba-test.pag-eng: 49/264
   Tatoeba-test.war-eng: 1512/9630
BLEU-scores:
   Tatoeba-test.akl-eng: 11.6
   Tatoeba-test.ceb-eng: 21.7
   Tatoeba-test.hil-eng: 17.6
   Tatoeba-test.ilo-eng: 36.6
   Tatoeba-test.multi-eng: 21.5
   Tatoeba-test.pag-eng: 27.5
   Tatoeba-test.war-eng: 17.3
chr-F-scores:
   Tatoeba-test.akl-eng: 0.321
   Tatoeba-test.ceb-eng: 0.393
   Tatoeba-test.hil-eng: 0.371
   Tatoeba-test.ilo-eng: 0.560
   Tatoeba-test.multi-eng: 0.391
   Tatoeba-test.pag-eng: 0.494
   Tatoeba-test.war-eng: 0.380