opus-2020-07-27.yml

release: inc-inc/opus-2020-07-27.zip
release-date: 2020-07-27
dataset-name: opus
modeltype: transformer
pre-processing: normalization + SentencePiece (spm32k,spm32k)
subwords:
   - source: spm32k
   - target: spm32k
subword-models:
   - source: source.spm
   - target: target.spm
source-languages:
   - asm
   - hin
   - mar
   - urd
target-languages:
   - asm
   - hin
   - mar
   - urd
use-target-labels:
   - >>asm<<
   - >>hin<<
   - >>mar<<
   - >>urd<<
test-data:
   - Tatoeba-test.asm-hin: 4/15
   - Tatoeba-test.hin-asm: 4/13
   - Tatoeba-test.hin-mar: 158/692
   - Tatoeba-test.hin-urd: 239/1591
   - Tatoeba-test.mar-hin: 158/860
   - Tatoeba-test.multi-multi: 802/4683
   - Tatoeba-test.urd-hin: 239/1512
BLEU-scores:
   - Tatoeba-test.asm-hin: 2.6
   - Tatoeba-test.hin-asm: 9.1
   - Tatoeba-test.hin-mar: 28.1
   - Tatoeba-test.hin-urd: 19.9
   - Tatoeba-test.mar-hin: 11.6
   - Tatoeba-test.multi-multi: 17.1
   - Tatoeba-test.urd-hin: 13.5
chr-F-scores:
   - Tatoeba-test.asm-hin: 0.231
   - Tatoeba-test.hin-asm: 0.262
   - Tatoeba-test.hin-mar: 0.548
   - Tatoeba-test.hin-urd: 0.508
   - Tatoeba-test.mar-hin: 0.466
   - Tatoeba-test.multi-multi: 0.464
   - Tatoeba-test.urd-hin: 0.377