神经机器翻译模型,用于将加泰罗尼亚语、奥克西唐语和西班牙语(cat+oci+spa)翻译为英语(en)。
该模型是 OPUS-MT project 的一部分,旨在使神经机器翻译模型广泛可用并对世界上许多语言进行访问。所有模型最初都是使用令人惊叹的 Marian NMT 框架进行训练的,该框架是用纯C++编写的高效NMT实现。这些模型已使用huggingface的transformers库转换为pyTorch。训练数据来自 OPUS ,并且训练流程使用了 OPUS-MT-train 的程序。
@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",
}
简单示例代码:
from transformers import MarianMTModel, MarianTokenizer
src_text = [
"¿Puedo hacerte una pregunta?",
"Toca algo de música."
]
model_name = "pytorch-models/opus-mt-tc-big-cat_oci_spa-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:
# Can I ask you a question?
# He plays some music.
您也可以使用transformers pipelines来使用OPUS-MT模型,例如:
from transformers import pipeline
pipe = pipeline("translation", model="Helsinki-NLP/opus-mt-tc-big-cat_oci_spa-en")
print(pipe("¿Puedo hacerte una pregunta?"))
# expected output: Can I ask you a question?
| langpair | testset | chr-F | BLEU | #sent | #words |
|---|---|---|---|---|---|
| cat-eng | tatoeba-test-v2021-08-07 | 0.72019 | 57.3 | 1631 | 12627 |
| spa-eng | tatoeba-test-v2021-08-07 | 0.76017 | 62.3 | 16583 | 138123 |
| cat-eng | flores101-devtest | 0.69572 | 45.4 | 1012 | 24721 |
| oci-eng | flores101-devtest | 0.63347 | 37.5 | 1012 | 24721 |
| spa-eng | flores101-devtest | 0.59696 | 29.9 | 1012 | 24721 |
| spa-eng | newssyscomb2009 | 0.57104 | 30.8 | 502 | 11818 |
| spa-eng | news-test2008 | 0.55440 | 27.9 | 2051 | 49380 |
| spa-eng | newstest2009 | 0.57153 | 30.2 | 2525 | 65399 |
| spa-eng | newstest2010 | 0.61890 | 36.8 | 2489 | 61711 |
| spa-eng | newstest2011 | 0.60278 | 34.7 | 3003 | 74681 |
| spa-eng | newstest2012 | 0.62760 | 38.6 | 3003 | 72812 |
| spa-eng | newstest2013 | 0.60994 | 35.3 | 3000 | 64505 |
| spa-eng | tico19-test | 0.74033 | 51.8 | 2100 | 56315 |
本工作得到 European Language Grid 的支持,作为 pilot project 2866 的一部分,受 FoTran project 的资助,该项目由欧洲研究理事会 (ERC) 在欧洲联盟的Horizon 2020研究与创新计划(赠款协议号 771113)下,以及由欧盟Horizon 2020研究与创新计划在赠款协议号 780069 下的 MeMAD project 提供资金支持。我们还感谢 CSC -- IT Center for Science 在芬兰提供的慷慨计算资源和IT基础设施。