Neural Machine Translation: A Beginner’s Guide

Neural Machine Translation: A Beginner’s Guide

Understanding NMT (Neural Machine Translation) can take a while. We know, we’ve been there. And now we are in a position to shed some light on this complex yet powerful technology.

There are several types of MT (Machine Translation) technology, each with different implications and features. Bare MT works differently than statistical MT and they both differ quite a bit from NMT. NMT is cutting edge technology in the language industry, and our choice for powering our Machine Translation services.

Behind the name

NMT’s name derives from a loose connection between how our brain’s neuron network functions and how this software is structured. This structure allows a system to be fed data and output results based on weights and biases. These two elements are the key leverages computational linguists have to tweak to obtain the best performing results: the highest quality translation.

A simple definition of NMT

It’s a set of incrementally improving rules written in software that, when applied to a source document, produce a target document. This target document contains the source’s meaning but expressed in the target language. NMT outperforms traditional MT in the fluidity of the output. But still, depending on the content it has translated, it is still possible for the reader to realize it comes from a machine. NMT engines do not perform like humans, so far. Or do they?

No machine or system yet understands the actual meaning of a text, but neural technology is equipped with word embedding algorithms and models that empower this technology to better represent the underlying structure of the text. Neural technology allows for the windows of context around each word to be wider than statistical MT does. It basically enables the NMT engine to leverage more word combinations from the source text, and previous translations too. And the process improves with every use. This is why we decided to make this technology the backbone of our MT offering. We believe that NMT is not just about getting the technology right. It’s also about having a strong partnership with customers that is characterized by collaboration and feedback.

Building trust with NMT

For some of our customers, go-to-market speed is key, for others, it is globally supporting their own client base. Across the board, there are enormous amounts of content (support, knowledge, social media, product information, and so on) waiting to be localized in multiple target markets. By enabling NMT solutions we are empowering our clients to deliver content fast, in multiple languages and regions.

Based on the findings of the Machine Translation Maturity Model white paper, by Valeria Cannavina, we have a clear understanding of what type of content, processes, and data is most suitable for NMT.  With Neural NMT still regarded as relatively new, many organizations overlook the importance of aligned procedures to guide their use of MT. At the same time, customers expect relevant information at their fingertips. We have helped multiple customers develop a model to succeed using NMT in their organizations.

Reaching an international readership, fast

Because NMT facilitates fast-flowing content, our customers’ global audiences have access to the most current digital information provided using our NMT solution. Aligning people and data review processes is especially critical to success with NMT. We understand that NMT requires adjustments by the user, as well as a process to identify the right type of content to leverage a maximum benefit.