opus2m-2020-08-01.yml

release: eng-bat/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:
   - lav
   - lit
   - ltg
   - prg
   - sgs
use-target-labels:
   - >>lav<<
   - >>lit<<
   - >>ltg<<
   - >>prg_Latn<<
   - >>sgs<<
test-data:
   - Tatoeba-test.eng-lav: 1631/7915
   - Tatoeba-test.eng-lit: 2500/11920
   - Tatoeba-test.eng-multi: 4396/21158
   - Tatoeba-test.eng-prg: 213/1243
   - Tatoeba-test.eng-sgs: 52/80
   - newsdev2017-enlv.eng-lav: 2003/34526
   - newsdev2019-enlt.eng-lit: 2000/32998
   - newstest2017-enlv.eng-lav: 2001/32754
   - newstest2019-enlt.eng-lit: 998/16593
BLEU-scores:
   - Tatoeba-test.eng-lav: 42.8
   - Tatoeba-test.eng-lit: 37.1
   - Tatoeba-test.eng-multi: 37.0
   - Tatoeba-test.eng-prg: 0.5
   - Tatoeba-test.eng-sgs: 4.1
   - newsdev2017-enlv.eng-lav: 24.0
   - newsdev2019-enlt.eng-lit: 20.9
   - newstest2017-enlv.eng-lav: 18.3
   - newstest2019-enlt.eng-lit: 13.6
chr-F-scores:
   - Tatoeba-test.eng-lav: 0.652
   - Tatoeba-test.eng-lit: 0.650
   - Tatoeba-test.eng-multi: 0.616
   - Tatoeba-test.eng-prg: 0.130
   - Tatoeba-test.eng-sgs: 0.178
   - newsdev2017-enlv.eng-lav: 0.546
   - newsdev2019-enlt.eng-lit: 0.533
   - newstest2017-enlv.eng-lav: 0.506
   - newstest2019-enlt.eng-lit: 0.466