Update October 9th 2019: This blog refers to adaptive/statistical machine translation, which has now been replaced by SDL’s powerful neural machine translation (NMT) solution. To learn more about NMT please click here.
If there is one thing which we can say with certainty about the world of translation, it is that it is always changing. Over the last ten years, we have witnessed an explosion in demand for translated content in new and increasingly technical formats and we have seen the industry transform its ways of working to handle this demand.
SDL has always been keen to be at the forefront of this change. As a translation technology provider, we have got to make sure we develop software that not only addresses the needs of the present but also anticipates the needs of the future. This is something of a balancing act!
Before talking about our plans for the future, it’s useful to talk a bit about how translation technology is used by translators every day, and how this has informed our development process.
How we are improving translation now
When a translator uses a CAT tool, they typically encounter three different scenarios each time they translate a new segment of text:
- Exact match: the translation memory (TM) identifies the new segment as a repetition of a previous one, and automatically reuses a stored translation. This is, traditionally, how the bulk of the productivity gain from CAT tools has been obtained.
- Fuzzy match: the TM identifies the new segment is similar to but not a perfect repetition of a previous segment. It reuses a stored translation, saving some time, but the translator needs to edit the translation more or less significantly before proceeding.
- No match: the TM cannot find any similarity between the new segment and the segments in its database. The segment is manually translated from scratch, or it can be pre-translated using a machine translation (MT) service and post-edited.
It is clear from this list that the bulk of the translator’s time is spent on the fuzzy and no match scenarios. So, naturally, improving this experience for translators is where we are focused. We aim to give translators more matches in total, higher quality fuzzy matches and useful translation suggestions as they work. This quest for the “best match" has become a guiding principle for us. And you can see this in SDL Trados Studio 2017 in the form of upLIFT and AdaptiveMT.
So what’s coming soon?
Firstly, you can expect more from upLIFT and AdaptiveMT. These technologies are fantastic cores which we are still optimizing and adding extra functionality to, which will arrive in updates to Studio 2017. You’ll have greater flexibility in how you work with upLIFT, compatibility with Chinese and Japanese language pairs and more.
On the machine translation front, SDL has recently stepped up efforts in the area of Neural Machine Translation (NMT) development as part of its ETS offering. This will be a fundamental advancement to our core MT engines – and it is not an understatement to say that what we are achieving significant quality improvements from just months of NMT development – which would have taken years or even been impossible with previous approaches to MT technology development.
We eventually plan to deliver the best of both worlds – AdaptiveMT’s learning capabilities on top of an NMT core – really pushing the boundaries of what MT can offer in terms of both quality and customization. In the short term, you’ll also see AdaptiveMT engines rolled out in new languages, supporting the growing number of translators using MT in no match scenarios and for translation suggestions at the fragment level in our AutoSuggest capability.
Secondly, we’re making more use of the cloud. More and more translators would like to be able to share TMs and terminology stored online, and using MT services. And this trend is only growing for all our audiences – be it groups of freelance translators, small LSPs without an on-premise server product such as SDL Trados GroupShare, or even corporate customers who would like to have less of an in-house technology footprint – so it’s important that desktop tools like Studio don’t limit translators when it comes to using the cloud. I often call this way of sharing assets in the cloud “the third way of working".
So far, we have supported two ways of sharing – file-based/local and on-premise/server. The third way of working will mean that we can fully democratize sharing projects, TMs, terminology and even AdaptiveMT engines in the cloud. This will mean that anyone will be able to work this way, not just users who happen to have the budget and IT to afford an on-premise solution. And all this integrated with the most powerful CAT desktop environment, SDL Trados Studio.
One subtle, yet significant, improvement currently in the making and coming to Studio 2017 in Service Release 1 (SR1) is LookAhead. The name is self-explanatory: instead of waiting until you move to the next segment before checking it against your translation resources, Studio will perform this check slightly further ahead.
For offline users, this is a nice user experience improvement – there will be virtually no delay anymore between confirming your current segment and getting matches for the next one. But for users who have resources in the cloud (which can take longer to access due to network latency and other factors), this will be a real timesaver that can change the way that users experience working with a remote server or cloud-based resources.
Finally, we’re continuing our work on ease of use. We’re continuing to clean Studio 2017’s user interface and user flows, making sure all of the new features can be configured and used intuitively. As part of this re-work, we’re constantly improving how Studio 2017 signposts users so that it’s easier to find information and support when you need it.
Our long term vision for the future
Ultimately, our vision is to ‘augment’ the translator’s workplace, how Common Sense Advisory has put it recently. With technologies like AdaptiveMT or upLIFT, we are definitely seeing very promising beginnings of this vision – but it is just a starting point.
We are working towards scenarios where users need to spend less and less time configuring the resources they use to ‘augment’ their productivity. Rather the technology should be clever enough – not least by using Artificial Intelligence and radically re-thought user experience designs – to work for users behind the scenes to always give them the best possible match for their current segment. This match could be a Neural MT match that has learned from the user through AdaptiveMT and so gives a high-quality starting point. Or it could be a fuzzy match from the TM that takes upLIFT fragments or even Neural MT fragments to give the user a match that will require very little adaptation – if at all.
At the heart of all this is our desire to help our users to be as productive as possible in a world where translation rates are constantly under pressure and expectations from work givers get higher and higher all the time. In this climate, it is key that the technology is there to help users to be more successful and be able to rise to the challenge that users are confronted with in the modern marketplace.
Hopefully this blog has provided some insight into how we are approaching the future of translation, what we aim to achieve and what that means for you as a user of our software. If you want to learn more about AdaptiveMT and upLIFT technology, click the links below.Find out more about AdaptiveMT Find out more about upLIFT