SDL Machine Translation: The future of Neural Machine Translation is here
Over the last 25 years, translation technology has progressed rapidly, with translators and linguists becoming privy to a more comprehensive set of translation-assisting tools than ever before. CAT tools with access to translation memories, termbases, and a lot of other resources have all been developed and continuously refined to support the day-to-day work of translation professionals. These tools have increased productivity for thousands of users worldwide by enabling faster work speeds, improving overall accuracy, and equipping translators with the ability to accept higher volumes of work.
The concept of machine translation has actually been around long before CAT tools began to surface, but historic incarnations of machine translation typically struggled with certain language pairs and overall translation quality. It is only within the course of the last few years that the development of Neural Machine Translation (NMT) has elevated the capability of machine translation to be used for content that, previously, it would never have been considered for.
But what is NMT exactly? Any how does it further enhance the experience of the human translator? Daniel Brockmann, Director of Product Management at SDL, explains everything in this eye-opening interview.
1.) So what is Neural Machine Translation (NMT) and why does it perform better than other Machine Translation (MT) models?
Neural Machine Translation (NMT) is a new approach to machine translation, where a computer uses deep learning to build an artificial neural network to teach itself how to translate between languages. It uses these neural networks to translate entire sentences without breaking them down into smaller parts.
This is far superior to previous versions of machine translation - rule-based, for example, which makes assumptions based on a set of pre-defined grammar or syntax rules, and statistical, which relies on a statistical algorithm to predict the correct sequence of words and phrases. Both these methods of machine translation have typically been restricted in their ability to provide fluent translations that need less post-editing.
In contrast, NMT, bolstered by powerful AI, attempts to determine the meaning of the source content and to replicate this in the target translation by sending it through various layers of “neurons” that work with each other to determine the most likely target language sentence. This approach consistently makes for a more fluent translation. The method used is not dissimilar to the AI-based approach used in image or speech recognition, for example, both of which have also made significant progress in the last few years.
2.) Does working with Neural Machine Translation replace working with a CAT tool?
Whilst NMT is a great new addition to the arsenal of tools that users have at their disposal to be more efficient in their translation work, ultimately nothing beats an exact translation memory match. So in my view, it’s important not to downplay the continued importance of this key asset prematurely.
By using NMT within the CAT tool environment, however, you get the additional benefit of having higher quality neural machine translation suggestions whilst working in a system you know and trust. You have your translation memories and termbases available to you whilst you work, and you have the option to use NMT for post-editing pre-translated content or as an auto suggest feature as per your preference, where your translation memory does not give you an exact or high fuzzy match.
For translators, opting to augment your CAT tool with NMT rather than completely replace it means this additional technology will slot seamlessly into the workflow you are comfortable and proficient in, without altering any of it other than very likely being more productive.
3.) How easy is it to set up and start using Neural Machine Translation in SDL Trados Studio 2019?
We've recently launched an update for Studio 2019 (Cumulative Update 4) which makes it very easy to access NMT.
Users may already be familiar with accessing SDL Machine Translation through the Language Cloud interface, and what we've done behind the scenes is simply replace the previous machine translation engine with the NMT engine which you can access there. It really is as simple as that!
4.) Why should people use SDL Machine Translation in Trados Studio 2019?
A key factor is that is our NMT is already waiting for you in SDL Trados Studio 2019, and it's free! Users have access to 500,000 characters per month for free with the option to upgrade to more as needed. When going to a higher volume, you can either use SDL Machine Translation online via the cloud, or it can be moved offline to your on-premise environment if preferred.
Secondly – we are the developers of both CAT and machine translation technology. This means you can get the best of both worlds from SDL – proven Trados translation technology on the one hand, and high quality neural machine translation from the SDL Machine Translation team who have worked on this technology for more than 15 years on the other hand. There is no third party source or storing of information, and your security is assured by our sophisticated encryption process. In short, you are in safe hands because SDL is not only the provider, but also the developer, of our NMT.
We are the only ones who can bring neural machine translation, translation memory and terminology, the three pillars of a CAT tool, together in one intelligent, unified offering. This opens up so many exciting prospects for what we can offer in the future.
We even recently won the “Best Machine Translation Solution” award from leading market intelligence organization, AI Breakthrough, which we are very proud of.
5.) What are the main benefits of SDL Machine Translation?
To put it simply, NMT increases productivity and quality across the entire supply chain. For language service providers (LSPs) and corporate organizations, it opens up the opportunity to have higher volumes of your content translated, and it’s now possible to start looking at different kinds of content too that may not have been considered for translation before at all.
As a translator myself, even in my current role of Product Manager, I still have to translate documents such as PowerPoint presentations from English to German or vice versa quite frequently, and I have found using SDL NMT to be a very nice enhancement to my translation process. It makes my way of working a lot more fluent as I always have access to something useful, whether it comes from the translation memory, the termbase, or now from neural machine translation.
In summary, the real beauty of this product is that it benefits all ways of working.