cometoid22-wmt23

A referenceless/quality-estimation metric for machine translation evaluation. This metric is created by using the knowledge distillation of wmt22-comet-da (a referece-based teacher). Refer to the publication for technical details.

Setup

Option 1: Install pymarian, aka python bindings to Marian

pip install pymarian

Option 2: Build marian binary, reference: https://marian-nmt.github.io/quickstart/

Usage

Pymarian

pymarian-eval -m checkpoints/marian.model.bin -v vocab.spm --like comet-qe  -s src.txt -t mt.out.txt

Marian

paste src.txt mt.out.txt | marian evaluate --quiet --model checkpoints/marian.model.bin --vocabs vocab.spm vocab.spm --width 4 --like comet-qe \
  --mini-batch 16 --maxi-batch 256 --max-length 512 --max-length-crop true --workspace 8000

More info at https://github.com/marian-nmt/wmt23-metrics

Reference

@inproceedings{gowda-etal-2023-cometoid,
    title = "Cometoid: Distilling Strong Reference-based Machine Translation Metrics into {E}ven Stronger Quality Estimation Metrics",
    author = "Gowda, Thamme  and
      Kocmi, Tom  and
      Junczys-Dowmunt, Marcin",
    editor = "Koehn, Philipp  and
      Haddon, Barry  and
      Kocmi, Tom  and
      Monz, Christof",
    booktitle = "Proceedings of the Eighth Conference on Machine Translation",
    month = dec,
    year = "2023",
    address = "Singapore",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.wmt-1.62",
    pages = "751--755",
}