opus4m-2020-08-12.yml

release: inc-eng/opus4m-2020-08-12.zip
release-date: 2020-08-12
dataset-name: opus4m
modeltype: transformer
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   source: spm32k
   target: spm32k
subword-models:
   source: source.spm
   target: target.spm
source-languages:
   - asm
   - awa
   - ben
   - bho
   - gom
   - guj
   - hif
   - hin
   - mai
   - mar
   - npi
   - ori
   - pan
   - pnb
   - rom
   - san
   - sin
   - snd
   - urd
target-languages:
   - eng
test-data:
   Tatoeba-test.asm-eng: 117/573
   Tatoeba-test.awa-eng: 279/1042
   Tatoeba-test.ben-eng: 2500/11321
   Tatoeba-test.bho-eng: 42/229
   Tatoeba-test.guj-eng: 154/767
   Tatoeba-test.hif-eng: 36/187
   Tatoeba-test.hin-eng: 5000/28315
   Tatoeba-test.kok-eng: 1/6
   Tatoeba-test.lah-eng: 32/162
   Tatoeba-test.mai-eng: 8/17
   Tatoeba-test.mar-eng: 10000/53713
   Tatoeba-test.multi-eng: 10000/53590
   Tatoeba-test.nep-eng: 115/389
   Tatoeba-test.ori-eng: 33/198
   Tatoeba-test.pan-eng: 87/514
   Tatoeba-test.rom-eng: 671/3765
   Tatoeba-test.san-eng: 144/504
   Tatoeba-test.sin-eng: 45/210
   Tatoeba-test.snd-eng: 4/15
   Tatoeba-test.urd-eng: 1663/10180
   newsdev2014.hin-eng: 520/9210
   newsdev2019-engu.guj-eng: 1998/36882
   newstest2014-hien.hin-eng: 2507/49394
   newstest2019-guen.guj-eng: 1016/15734
BLEU-scores:
   Tatoeba-test.asm-eng: 19.8
   Tatoeba-test.awa-eng: 8.8
   Tatoeba-test.ben-eng: 45.1
   Tatoeba-test.bho-eng: 25.7
   Tatoeba-test.guj-eng: 21.8
   Tatoeba-test.hif-eng: 9.0
   Tatoeba-test.hin-eng: 39.2
   Tatoeba-test.kok-eng: 1.8
   Tatoeba-test.lah-eng: 17.5
   Tatoeba-test.mai-eng: 53.2
   Tatoeba-test.mar-eng: 26.6
   Tatoeba-test.multi-eng: 30.0
   Tatoeba-test.nep-eng: 3.8
   Tatoeba-test.ori-eng: 5.8
   Tatoeba-test.pan-eng: 17.3
   Tatoeba-test.rom-eng: 1.8
   Tatoeba-test.san-eng: 4.8
   Tatoeba-test.sin-eng: 32.0
   Tatoeba-test.snd-eng: 38.5
   Tatoeba-test.urd-eng: 26.6
   newsdev2014.hin-eng: 9.2
   newsdev2019-engu.guj-eng: 10.1
   newstest2014-hien.hin-eng: 13.8
   newstest2019-guen.guj-eng: 6.9
chr-F-scores:
   Tatoeba-test.asm-eng: 0.382
   Tatoeba-test.awa-eng: 0.234
   Tatoeba-test.ben-eng: 0.601
   Tatoeba-test.bho-eng: 0.411
   Tatoeba-test.guj-eng: 0.386
   Tatoeba-test.hif-eng: 0.288
   Tatoeba-test.hin-eng: 0.570
   Tatoeba-test.kok-eng: 0.147
   Tatoeba-test.lah-eng: 0.315
   Tatoeba-test.mai-eng: 0.713
   Tatoeba-test.mar-eng: 0.504
   Tatoeba-test.multi-eng: 0.510
   Tatoeba-test.nep-eng: 0.206
   Tatoeba-test.ori-eng: 0.229
   Tatoeba-test.pan-eng: 0.370
   Tatoeba-test.rom-eng: 0.172
   Tatoeba-test.san-eng: 0.173
   Tatoeba-test.sin-eng: 0.525
   Tatoeba-test.snd-eng: 0.500
   Tatoeba-test.urd-eng: 0.468
   newsdev2014.hin-eng: 0.350
   newsdev2019-engu.guj-eng: 0.339
   newstest2014-hien.hin-eng: 0.410
   newstest2019-guen.guj-eng: 0.297