opus-mt-tc-big-fr-en

Neural machine translation model for translating from French (fr) to English (en).

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.

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",
}

Model info

Release: 2022-03-09
source language(s): fra
target language(s): eng
model: transformer-big
data: opusTCv20210807+bt (source)
tokenization: SentencePiece (spm32k,spm32k)
original model: opusTCv20210807+bt_transformer-big_2022-03-09.zip
more information released models: OPUS-MT fra-eng README

Usage

A short example code:

from transformers import MarianMTModel, MarianTokenizer

src_text = [
    "J'ai adoré l'Angleterre.",
    "C'était la seule chose à faire."
]

model_name = "pytorch-models/opus-mt-tc-big-fr-en"
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:
#     I loved England.
#     It was the only thing to do.

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-fr-en")
print(pipe("J'ai adoré l'Angleterre."))

# expected output: I loved England.

Benchmarks

test set translations: opusTCv20210807+bt_transformer-big_2022-03-09.test.txt
test set scores: opusTCv20210807+bt_transformer-big_2022-03-09.eval.txt
benchmark results: benchmark_results.txt
benchmark output: benchmark_translations.zip

langpair	testset	chr-F	BLEU	#sent	#words
fra-eng	tatoeba-test-v2021-08-07	0.73772	59.8	12681	101754
fra-eng	flores101-devtest	0.69350	46.0	1012	24721
fra-eng	multi30k_test_2016_flickr	0.68005	49.7	1000	12955
fra-eng	multi30k_test_2017_flickr	0.70596	52.0	1000	11374
fra-eng	multi30k_test_2017_mscoco	0.69356	50.6	461	5231
fra-eng	multi30k_test_2018_flickr	0.65751	44.9	1071	14689
fra-eng	newsdiscussdev2015	0.59008	34.4	1500	27759
fra-eng	newsdiscusstest2015	0.62603	40.2	1500	26982
fra-eng	newssyscomb2009	0.57488	31.1	502	11818
fra-eng	news-test2008	0.54316	26.5	2051	49380
fra-eng	newstest2009	0.56959	30.4	2525	65399
fra-eng	newstest2010	0.59561	33.4	2489	61711
fra-eng	newstest2011	0.60271	33.8	3003	74681
fra-eng	newstest2012	0.59507	33.6	3003	72812
fra-eng	newstest2013	0.59691	34.8	3000	64505
fra-eng	newstest2014	0.64533	39.4	3003	70708
fra-eng	tico19-test	0.63326	41.3	2100	56323

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: 3405783
port time: Wed Apr 13 19:02:28 EEST 2022
port machine: LM0-400-22516.local

Helsinki-NLP
/

opus-mt-tc-big-fr-en