unbabel-comet 0.0.6.post1

High-quality Machine Translation Evaluation

Note: This is a Pre-Release Version. We are currently working on results for the WMT2020 shared task and will likely update the repository in the beginning of October (after the shared task results).

Quick Installation

We recommend python 3.6 to run COMET.

Detailed usage examples and instructions can be found in the Full Documentation.

Simple installation from PyPI

pip install unbabel-comet

To develop locally:

git clone https://github.com/Unbabel/COMET
pip install -r requirements.txt
pip install -e .

Scoring MT outputs:

Via Bash:

Examples from WMT20:

echo -e "Dem Feuer konnte Einhalt geboten werden\nSchulen und Kindergärten wurden eröffnet." >> src.de
echo -e "The fire could be stopped\nSchools and kindergartens were open" >> hyp.en
echo -e "They were able to control the fire.\nSchools and kindergartens opened" >> ref.en

comet score -s src.de -h hyp.en -r ref.en

You can export your results to a JSON file using the --to_json flag and select another model/metric with --model.

comet score -s src.de -h hyp.en -r ref.en --model wmt-large-hter-estimator --to_json segments.json

Via Python:

from comet.models import download_model
model = download_model("wmt-large-da-estimator-1719")
data = [
    {
        "src": "Dem Feuer konnte Einhalt geboten werden",
        "mt": "The fire could be stopped",
        "ref": "They were able to control the fire."
    },
    {
        "src": "Schulen und Kindergärten wurden eröffnet.",
        "mt": "Schools and kindergartens were open",
        "ref": "Schools and kindergartens opened"
    }
]
model.predict(data)

Simple Pythonic way to convert list or segments to model inputs:

source = ["Dem Feuer konnte Einhalt geboten werden", "Schulen und Kindergärten wurden eröffnet."]
hypothesis = ["The fire could be stopped", "Schools and kindergartens were open"]
reference = ["They were able to control the fire.", "Schools and kindergartens opened"]

data = {"src": source, "mt": hypothesis, "ref": reference}
data = [dict(zip(data, t)) for t in zip(*data.values())]

model.predict(data)

Note: Using the python interface you will get a list of segment-level scores. You can obtain the corpus-level score by averaging the segment-level scores

Model Zoo:

Model	Description
↑wmt-large-da-estimator-1719	RECOMMENDED: Estimator model build on top of XLM-R (large) trained on DA from WMT17, WMT18 and WMT19
↑wmt-base-da-estimator-1719	Estimator model build on top of XLM-R (base) trained on DA from WMT17, WMT18 and WMT19
↓wmt-large-hter-estimator	Estimator model build on top of XLM-R (large) trained to regress on HTER.
↓wmt-base-hter-estimator	Estimator model build on top of XLM-R (base) trained to regress on HTER.
↑emnlp-base-da-ranker	Translation ranking model that uses XLM-R to encode sentences. This model was trained with WMT17 and WMT18 Direct Assessments Relative Ranks (DARR).

QE-as-a-metric:

Model	Description
wmt-large-qe-estimator-1719	Quality Estimator model build on top of XLM-R (large) trained on DA from WMT17, WMT18 and WMT19.

Train your own Metric:

Instead of using pretrained models your can train your own model with the following command:

comet train -f {config_file_path}.yaml

Supported encoders:

• Learning Joint Multilingual Sentence Representations with Neural Machine Translation
• Massively Multilingual Sentence Embeddings for Zero-Shot Cross-Lingual Transfer and Beyond
• BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
• XLM-R: Unsupervised Cross-lingual Representation Learning at Scale

Tensorboard:

Launch tensorboard with:

tensorboard --logdir="experiments/"

Download Command:

To download public available corpora to train your new models you can use the download command. For example to download the APEQUEST HTER corpus just run the following command:

comet download -d apequest --saving_path data/

unittest:

pip install coverage

In order to run the toolkit tests you must run the following command:

coverage run --source=comet -m unittest discover
coverage report -m

Publications

@inproceedings{rei-etal-2020-comet,
    title = "{COMET}: A Neural Framework for {MT} Evaluation",
    author = "Rei, Ricardo  and
      Stewart, Craig  and
      Farinha, Ana C  and
      Lavie, Alon",
    booktitle = "Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/2020.emnlp-main.213",
    pages = "2685--2702",
}

@inproceedings{rei-EtAl:2020:WMT,
  author    = {Rei, Ricardo  and  Stewart, Craig  and  Farinha, Ana C  and  Lavie, Alon},
  title     = {Unbabel's Participation in the WMT20 Metrics Shared Task},
  booktitle      = {Proceedings of the Fifth Conference on Machine Translation},
  month          = {November},
  year           = {2020},
  address        = {Online},
  publisher      = {Association for Computational Linguistics},
  pages     = {909--918},
}

@inproceedings{stewart-etal-2020-comet,
    title = "{COMET} - Deploying a New State-of-the-art {MT} Evaluation Metric in Production",
    author = "Stewart, Craig  and
      Rei, Ricardo  and
      Farinha, Catarina  and
      Lavie, Alon",
    booktitle = "Proceedings of the 14th Conference of the Association for Machine Translation in the Americas (Volume 2: User Track)",
    month = oct,
    year = "2020",
    address = "Virtual",
    publisher = "Association for Machine Translation in the Americas",
    url = "https://www.aclweb.org/anthology/2020.amta-user.4",
    pages = "78--109",
}