Fork me on GitHub

Вопросы и ответы


Why neural machine translation (NMT)?

We care about neural machine translation for several reasons.

1) Results show that NMT produces automatic translations that are significantly preferred by humans to other machine translation outputs. This has led several companies to switch NMT-based translation systems.

2) Similar methods (often called seq2seq) are also effective for many other NLP and language-related applications such as dialogue, image captioning, and summarization. A recent talk from HarvardNLP describing some of these recent advances is available here.

3) NMT has been used as a representative application of the recent success of deep learning-based artificial intelligence. For instance a recent NYT magazine cover story focused on Google’s NMT system.

Where can I go to learn about NMT?

We recommend starting with the ACL’16 NMT tutorial produced by researchers at Stanford and NYU. The tutorial also includes a detailed bibliography describing work in the field.

What do I need to train an NMT model?

You just need two files: a source file and a target file. Each with one sentence per line with words space separated. These files can come from standard free translation corpora such a WMT, or it can be any other sources you want to train from.

Installation / Setup

What type of computer do I need to train with?

While in theory you can train on any machine; in practice for all but trivally small data sets you will need a GPU that supports CUDA if you want training to finish in a reasonable amount of time. For medium-size models you will need at least 4GB; for full-size state-of-the-art models 8-12GB is recommend.

How can I trust that the current version of OpenNMT is working?

We run a daily integration test of the system on EC2 to check that it is functioning properly. The logs of the system are publicly available in the integration test repo.


How can I replicate the full-scale NMT translation results?

We have posted a complete tutorial for training a German-to-English translation system on standard data.

Are there pretrained translation models that I can try?

There are several different pretrained models available on the models page.

Where can I get training data for translation from X-to-X?

Try the OPUS Project. An open-source collection of parallel corpora. After stripping XML tags, you should be able to use the raw files directly in OpenNMT.

I am interested in other seq2seq-like problems such as summarization, dialogue, tree-generation. Can OpenNMT work for these?

Yes. OpenNMT is a general-purpose attention-based seq2seq system. There is very little code that is translation specific, and so it should be effective for many of these applications.

For the case of summarization, OpenNMT has been shown to be more effective than neural systems like NAMAS, and will be supported going forward. See the models page for a pretrained summarization system on the Gigaword dataset.

I am interested in variants of seq2seq such as image-to-sequence generation. Can OpenNMT work for these?

Partially. OpenNMT can be extended with additional Torch neural networks, and a bit of coding. As an example, we have implemented a relatively general-purpose im2text system, with a small amount of additional code. Feel free to use this as a model for extending OpenNMT.

Additional Features

Is the OpenNMT demo server available?

Unfortunately we cannot open-source the demo server. However you can run OpenNMT in server mode and we are developing aweb interface. Contributions are welcome.


Can I send a pull request?

That would be great. Just be sure to read the before you submit.

Where do the developers talk?

We are normally in our Gitter channel. If you want to contribute come chat. We are friendly.

Feature X would make OpenNMT even better!

That’s not really a question, but we would love to add that feature! The quickest way to get it into OpenNMT is to send us a pull request on GitHub. Just fork the repo, make the change, and click send pull request.

But I don’t know how to code…

Okay, that’s fine, we can help implement it. Go to our forum and add a feature request. We will get to it soon.