Options for Machine Translation on AWS ?

Here’s another question I got recently: “What are the different options for Machine Translation on AWS?”

You have three options. Let’s go through each one of them.

1 — Use Amazon Translate

Amazon Translate is a high-level AI service based on Deep Learning technology. It delivers fast, high-quality, and affordable language translation.

At the time of writing, it supports 12 language pairs (more coming): English-Arabic, English-Chinese (simplified), English-French, English-German, English-Portuguese, English-Spanish.

If you want to see Amazon Translate in action, please refer to this previous post.

2 — Build a custom model on managed infrastructure with Amazon SageMaker

Amazon SageMaker is a fully-managed service that enables developers and data scientists to quickly and easily build, train, and deploy Machine Learning models at scale.

One of the main features of SageMaker is its collection of built-in Amazon-implemented algorithms. One of these algorithms is ‘seq2seq’, a popular choice to build custom Machine Translation models. In particular, this could help you train a model for a language pair not yet available in Amazon Translate.

If you want to see ‘seq2seq’ in action, please refer to this Jupyter notebook that shows you how to train a model translating English to German. It is also available on Sagemaker notebook instances under sample-notebooks/introduction_to_amazon_algorithms/seq2seq_translation_en-de.

3 — Build a custom model on any infrastructure with Sockeye

Sockeye is an open source project that lets you train Machine Translation models. Sockeye also uses the ‘seq2seq’ algorithm and is based on Apache MXNet, a open source library for Deep Learning.

With Sockeye, you can train custom models on any infrastructure. Using GPU instances from the P3 family and the Deep Learning AMI on AWS would make sense, but obviously you can use it anywhere else, including on your own servers.

To see Sockeye in action, please refer to this detailed tutorial on translating German to English.

As you can see, the choice is yours!

Have another question? Please ping me on Twitter.