opus-mt-tc-big-de-gmq

Table of Contents

• Model Details
• Uses
• Risks, Limitations and Biases
• How to Get Started With the Model
• Training
• Evaluation
• Citation Information
• Acknowledgements

Model Details

Neural machine translation model for translating from German (de) to North Germanic languages (gmq).

This model is part of the OPUS-MT project, an effort to make neural machine translation models widely available and accessible for many languages in the world. All models are originally trained using the amazing framework of Marian NMT, an efficient NMT implementation written in pure C++. The models have been converted to pyTorch using the transformers library by huggingface. Training data is taken from OPUS and training pipelines use the procedures of OPUS-MT-train. Model Description:

• Developed by: Language Technology Research Group at the University of Helsinki
• Model Type: Translation (transformer-big)
• Release: 2022-07-25
• License: CC-BY-4.0
• Language(s):
  • Source Language(s): deu
  • Target Language(s): dan isl nno nob nor swe
  • Language Pair(s): deu-dan deu-isl deu-nob deu-swe
  • Valid Target Language Labels: >>dan<< >>fao<< >>isl<< >>jut<< >>nno<< >>nob<< >>non<< >>nrn<< >>ovd<< >>qer<< >>rmg<< >>swe<<
• Original Model: opusTCv20210807_transformer-big_2022-07-25.zip
• Resources for more information:
  • OPUS-MT-train GitHub Repo
  • More information about released models for this language pair: OPUS-MT deu-gmq README
  • More information about MarianNMT models in the transformers library
  • [Tatoeba Translation Challenge](https://github.com/Helsinki-NLP/Tatoeba-Challenge/

This is a multilingual translation model with multiple target languages. A sentence initial language token is required in the form of >>id<< (id = valid target language ID), e.g. >>dan<<

Uses

This model can be used for translation and text-to-text generation.

Risks, Limitations and Biases

CONTENT WARNING: Readers should be aware that the model is trained on various public data sets that may contain content that is disturbing, offensive, and can propagate historical and current stereotypes.

Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)).

How to Get Started With the Model

A short example code:

from transformers import MarianMTModel, MarianTokenizer

src_text = [
    ">>dan<< Ich hätte fast meinen Pass vergessen.",
    ">>dan<< Dieses Fenster hier ist schusssicher."
]

model_name = "pytorch-models/opus-mt-tc-big-de-gmq"
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)
translated = model.generate(**tokenizer(src_text, return_tensors="pt", padding=True))

for t in translated:
    print( tokenizer.decode(t, skip_special_tokens=True) )

# expected output:
#     Jeg havde næsten glemt mit pas.
#     Dette vindue er skudsikkert.

You can also use OPUS-MT models with the transformers pipelines, for example:

from transformers import pipeline
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-de-gmq")
print(pipe(">>dan<< Ich hätte fast meinen Pass vergessen."))

# expected output: Jeg havde næsten glemt mit pas.

Training

• Data: opusTCv20210807 (source)
• Pre-processing: SentencePiece (spm32k,spm32k)
• Model Type: transformer-big
• Original MarianNMT Model: opusTCv20210807_transformer-big_2022-07-25.zip
• Training Scripts: GitHub Repo

Evaluation

• test set translations: opusTCv20210807_transformer-big_2022-07-25.test.txt
• test set scores: opusTCv20210807_transformer-big_2022-07-25.eval.txt
• benchmark results: benchmark_results.txt
• benchmark output: benchmark_translations.zip

langpair	testset	chr-F	BLEU	#sent	#words
deu-dan	tatoeba-test-v2021-08-07	0.74306	58.7	9998	74644
deu-isl	tatoeba-test-v2021-08-07	0.65180	47.1	969	5951
deu-nob	tatoeba-test-v2021-08-07	0.71062	52.5	3525	31978
deu-swe	tatoeba-test-v2021-08-07	0.72658	58.3	3410	22701
deu-dan	flores101-devtest	0.62363	35.6	1012	24638
deu-isl	flores101-devtest	0.48691	20.6	1012	22834
deu-nob	flores101-devtest	0.55765	25.3	1012	23873
deu-swe	flores101-devtest	0.62323	34.7	1012	23121

Citation Information

• Publications: OPUS-MT – Building open translation services for the World and The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT (Please, cite if you use this model.)

@inproceedings{tiedemann-thottingal-2020-opus,
    title = "{OPUS}-{MT} {--} Building open translation services for the World",
    author = {Tiedemann, J{\"o}rg  and Thottingal, Santhosh},
    booktitle = "Proceedings of the 22nd Annual Conference of the European Association for Machine Translation",
    month = nov,
    year = "2020",
    address = "Lisboa, Portugal",
    publisher = "European Association for Machine Translation",
    url = "https://aclanthology.org/2020.eamt-1.61",
    pages = "479--480",
}

@inproceedings{tiedemann-2020-tatoeba,
    title = "The Tatoeba Translation Challenge {--} Realistic Data Sets for Low Resource and Multilingual {MT}",
    author = {Tiedemann, J{\"o}rg},
    booktitle = "Proceedings of the Fifth Conference on Machine Translation",
    month = nov,
    year = "2020",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2020.wmt-1.139",
    pages = "1174--1182",
}

Acknowledgements

The work is supported by the European Language Grid as pilot project 2866, by the FoTran project, funded by the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 771113), and the MeMAD project, funded by the European Union’s Horizon 2020 Research and Innovation Programme under grant agreement No 780069. We are also grateful for the generous computational resources and IT infrastructure provided by CSC -- IT Center for Science, Finland.

Model conversion info

• transformers version: 4.16.2
• OPUS-MT git hash: 8b9f0b0
• port time: Sat Aug 13 00:07:19 EEST 2022
• port machine: LM0-400-22516.local