README.md

1.1.0 of 30 May 2022

Changelog

• use moses EMS (Experiment Management System) to optimize the translation model
• delete the pre-generated models from zmifanva repository
• split zmifanva corpus on train, validation and test
• generate moses model, containerize it and push to docker hub

Announce on Reddit

Run

docker-compose up
xdg-open http://localhost:6543

Source

• lojban-mt: tag 1.1.0
• zmifanva: tag 1.1.0
• generating moses model for zmifanva

Evaluation

English to Lojban

test

BLEU = 68.35 100.0/79.3/57.1/48.1 (BP = 1.000 ratio = 1.034 hyp_len = 30 ref_len = 29)
chrF2 = 78.05
TER = 113.63

train

BLEU = 89.01 96.9/90.3/86.7/82.8 (BP = 1.000 ratio = 1.000 hyp_len = 32 ref_len = 32)
chrF2 = 94.50
TER = 30.43

validation

BLEU = 68.34 100.0/81.8/60.0/44.4 (BP = 1.000 ratio = 1.000 hyp_len = 12 ref_len = 12)
chrF2 = 86.94
TER = 14.29

Lojban to English

test

BLEU = 45.85 77.1/67.6/42.4/28.1 (BP = 0.918 ratio = 0.921 hyp_len = 35 ref_len = 38)
chrF2 = 63.27
TER = 237.11

train

BLEU = 49.58 78.6/55.6/38.5/36.0 (BP = 1.000 ratio = 1.000 hyp_len = 28 ref_len = 28)
chrF2 = 48.09
TER = 224.65

validation

BLEU = 17.24 58.3/27.3/10.0/5.6 (BP = 1.000 ratio = 1.000 hyp_len = 12 ref_len = 12)
chrF2 = 23.99
TER = 119.18