OPUS-MT-train/models/ht-en
2020-05-29 14:43:36 +03:00
..
README.md fixed a bug in eval-testsets 2020-05-29 14:43:36 +03:00

opus-2020-01-09.zip

Benchmarks

testset BLEU chr-F
JW300.ht.en 37.5 0.542
Tatoeba.ht.en 57.0 0.689

opus-2020-05-23.zip

Training data: opus+bt

  • ht-en: QED (4643) Tatoeba (45) Ubuntu (225) wikimedia (83)
  • ht-en: total size = 4996
  • unused dev/test data is added to training data
  • total size (opus+bt): 204693

Validation data

  • en-ht: JW300

  • devset = top 2500 lines of JW300.src.shuffled!

  • testset = next 2500 lines of JW300.src.shuffled!

  • remaining lines are added to traindata

Benchmarks

testset BLEU chr-F
JW300.ht.en 36.8 0.545