The interesting thing about AdaptiveMT, often also referred to as “self-learning" machine translation (MT), is that it is not actually a new technology to us at SDL. We have in one way or another been working with AdaptiveMT technology since 2005 when we filed our first self-learning machine translation patent. With the release of SDL Trados Studio 2017, AdaptiveMT is a standard service, so why is now the right time to introduce this technology?
What our research told us
Earlier this year SDL conducted the largest technology research initiative for our industry, Translation Technology Insights, with nearly 3,000 respondents. From this research, unsurprisingly, productivity was a key concern for people.
A key trend raised was how can we continue to cope with the ever increasing demands to turn around translations quicker than ever? So when we asked you about the next productivity boost and where might it come from, machine translation was clearly identified.
Given the controversy surrounding MT this was a little bit of a surprise for us. Despite the still relatively low adoption level of machine translation, MT usage is projected to rise significantly over the next 5 years. In fact machine translation has already made some credible headway in recent years.
Overall, 44% of respondents to our research believe that MT is driving the future of the industry.
Interestingly, of those that already use MT, 59% see MT having a positive impact for the future.
Even though machine translation has made some big strides there are still concerns and issues that have needed addressing, the most common being post-editing.
Post-edits are always ignored as the MT engine, even if it has been trained, is “static" and cannot adapt to any changes the translator makes. In addition to reducing productivity, this causes frustration for the user as the same mistakes are repeated again and again by the engines.
AdaptiveMT, on the other hand, addresses these issues and represents an exciting innovation in the field of machine translation, but what exactly is it and how does it help the life of a translator?
How AdaptiveMT can benefit you
A self-learning MT engine unlike traditional MT adapts in real time to the terminology and style of the translator, based on each individual post-edited segment that is sent back to the engine.
While speeding up the editing of new machine translated content, this also addresses one of the key concerns for using MT in a translation productivity environment – having to correct the same mistakes over and over again. The machine learns from the translator and thus errors are reduced significantly over time.
Any post-edit done by a translator is taken on board by SDL’s MT engine as it is based on an adaptive model.
What is also important here is that the translator does not have to learn something new or change their behavior; the post editing process remains the same, it is just active, meaning it remembers what you have done. In effect you have your own personalized machine translation engine.
This is potentially a paradigm shift for the translation industry, as for the first time users can update a machine translation engine as they go along with new suggestions, which is not that dissimilar to updating a translation memory. This in turn means that MT technology may well be accepted by a far bigger audience than previously, as one of the biggest frustrations will start being addressed by this technology shift. Ultimately, with this shift, machine translation and translation memory technology start converging.
Configuration in SDL Trados Studio 2017
To work with an adaptive engine, all the user needs to do is create an engine and plug it into SDL Trados Studio 2017:
Creating an adaptive machine translation engine in SDL Trados Studio 2017
What’s more, for the first time, users can indicate that they want to update a machine translation engine. By design, this was not possible with static MT engines:
Selecting to update an adaptive machine translation engine
We are very excited at SDL to be delivering this innovative technology to the market but it is just the start and should also be seen as a step towards our BEST Match principle – the convergence of translation memory and machine translation I alluded to earlier.
Simply put, our aim is that whenever a translator translates they are never left with an empty segment, they should always have access to a suggestion from SDL based on their assets. AdaptiveMT combined with new TM upLIFT technology in SDL Trados Studio 2017 as our Director of Product Management, Daniel Brockmann, would say is “one small click for a translator, a big step for the industry."