opus2m-2020-08-01.yml

release: trk-eng/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:
   - aze
   - bak
   - chv
   - crh
   - kaz
   - kir
   - kjh
   - kum
   - ota
   - sah
   - tat
   - tuk
   - tur
   - tyv
   - uig
   - uzb
target-languages:
   - eng
test-data:
   - Tatoeba-test.aze-eng: 2659/13295
   - Tatoeba-test.bak-eng: 39/173
   - Tatoeba-test.chv-eng: 333/1876
   - Tatoeba-test.crh-eng: 22/92
   - Tatoeba-test.kaz-eng: 397/2176
   - Tatoeba-test.kir-eng: 118/583
   - Tatoeba-test.kjh-eng: 17/60
   - Tatoeba-test.kum-eng: 8/41
   - Tatoeba-test.multi-eng: 3053/15436
   - Tatoeba-test.ota-eng: 678/3612
   - Tatoeba-test.sah-eng: 39/183
   - Tatoeba-test.tat-eng: 1451/9447
   - Tatoeba-test.tuk-eng: 2500/16260
   - Tatoeba-test.tur-eng: 10000/66526
   - Tatoeba-test.tyv-eng: 5/20
   - Tatoeba-test.uig-eng: 3024/19231
   - Tatoeba-test.uzb-eng: 457/2184
   - newsdev2016-entr.tur-eng: 1001/19474
   - newstest2016-entr.tur-eng: 3000/58396
   - newstest2017-entr.tur-eng: 3007/59741
   - newstest2018-entr.tur-eng: 3000/60232
BLEU-scores:
   - Tatoeba-test.aze-eng: 15.1
   - Tatoeba-test.bak-eng: 3.3
   - Tatoeba-test.chv-eng: 1.3
   - Tatoeba-test.crh-eng: 10.8
   - Tatoeba-test.kaz-eng: 9.6
   - Tatoeba-test.kir-eng: 15.3
   - Tatoeba-test.kjh-eng: 1.8
   - Tatoeba-test.kum-eng: 16.1
   - Tatoeba-test.multi-eng: 12.0
   - Tatoeba-test.ota-eng: 2.0
   - Tatoeba-test.sah-eng: 0.7
   - Tatoeba-test.tat-eng: 4.0
   - Tatoeba-test.tuk-eng: 5.5
   - Tatoeba-test.tur-eng: 26.8
   - Tatoeba-test.tyv-eng: 1.3
   - Tatoeba-test.uig-eng: 0.2
   - Tatoeba-test.uzb-eng: 4.6
   - newsdev2016-entr.tur-eng: 5.0
   - newstest2016-entr.tur-eng: 3.7
   - newstest2017-entr.tur-eng: 3.7
   - newstest2018-entr.tur-eng: 4.1
chr-F-scores:
   - Tatoeba-test.aze-eng: 0.330
   - Tatoeba-test.bak-eng: 0.185
   - Tatoeba-test.chv-eng: 0.161
   - Tatoeba-test.crh-eng: 0.325
   - Tatoeba-test.kaz-eng: 0.264
   - Tatoeba-test.kir-eng: 0.328
   - Tatoeba-test.kjh-eng: 0.121
   - Tatoeba-test.kum-eng: 0.277
   - Tatoeba-test.multi-eng: 0.304
   - Tatoeba-test.ota-eng: 0.149
   - Tatoeba-test.sah-eng: 0.140
   - Tatoeba-test.tat-eng: 0.215
   - Tatoeba-test.tuk-eng: 0.243
   - Tatoeba-test.tur-eng: 0.443
   - Tatoeba-test.tyv-eng: 0.111
   - Tatoeba-test.uig-eng: 0.111
   - Tatoeba-test.uzb-eng: 0.195
   - newsdev2016-entr.tur-eng: 0.242
   - newstest2016-entr.tur-eng: 0.231
   - newstest2017-entr.tur-eng: 0.229
   - newstest2018-entr.tur-eng: 0.230