SDL AutoSuggest 2.0 & LC: Working Together For Your Translations

Update October 9th 2019: This blog refers to adaptive/statistical machine translation, which has now been replaced by SDL’s powerful neural machine translation (NMT) solution.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.

It feels like only yesterday, although it was over 6 years ago, that we released SDL Trados Studio 2009, a landmark which heralded the next generation in translation memory.

A key part of this revolution in translation memory technology was my favourite feature – SDL AutoSuggest. It introduced a very innovative way to deliver suggestions as you type (unsurprisingly!), using previous translations. It came in handy when a traditional translation memory match was not available but part of a sentence had already been translated.

These suggestions are incredibly smart and also include terms, formatting and tags. There was one downside: you needed 10,000 translation memory segments to be able to create an AutoSuggest dictionary that could be leveraged for your translations. This resulted in very good accuracy in the suggestions but it meant not everybody could benefit from this fantastic feature.

Fast forward to the present and AutoSuggest has received some fundamental and exciting changes in SDL Trados Studio 2015. With the latest release of Studio, SDL AutoSuggest has turned into an incredibly versatile and powerful resource to provide possible translations and help you translate faster than ever before.

In my preparations for demonstrating Studio 2015 and the new AutoSuggest 2.0 to customers I find it very useful to translate some content myself, in this case some of our own marketing materials. I have deliberately used a translation memory which was out of date and did not include the last three years of translations. What I found was the results were very impressive and I really noticed how much faster I could translate some sentences.

With this in mind, I focused on two areas to see how SDL AutoSuggest 2.0 acted in the real world:

  • How it could leverage translation memories without creating an AutoSuggest dictionary (which you still can do, by the way)
  • More intriguingly, how the combination of Machine Translation and AutoSuggest would work together.

I was pleasantly surprised by the leveraging of the translation memory and stunned by how AutoSuggest and Machine Translation work together.

SDL AutoSuggest and Language Cloud Concordance

Looking at my first point – suggestions like concordance and fuzzy matches were suggested to me. I could choose to use the suggestions or ignore them – and it is possible to switch each component on or off (100%, Fuzzy and concordance). I felt I saved a lot of time spent typing, meaning I had better consistency in the translation, as I could easily leverage concordance and could easily fix a sentences’ fragments to complete the translation quickly and efficiently.

The game changer which surprised me the most was when I connected AutoSuggest to Machine Translation (MT). I used, of course, SDL LC. My language pair, i.e. English to Italian, seems to be well suited as the Machine Translation quality seemed pretty good to me.

Using AutoSuggest with Machine Translation introduces a very different way of working and to be honest, at times I have struggled with Machine Translation in the past. I have not been formally trained on post-editing, which lead me to more often than not deleting an entire MT translated sentence to change two words endings, leaving me feeling a little frustrated (practicing what I preach at some post-editing webinars did not really come into play!)

However in Studio 2015 I am getting shorter suggestions … more modular suggestions in fact. As I type and I can easily decide what to use and what to discard. It feels more natural to me, more fluid as a process.

This feels like a real transformation in the process that can bring a better way to work with MT. In a way no more post-editing: just translating normally but with useful extra suggestions. Many translators have been used to leveraging translation memory, terminology and more recently AutoSuggest dictionaries. This latest innovation added one more source, which happens to be Machine Translation. The way it is presented and implemented is what makes it much more usable and effective. I know it might not be for everyone, especially if your language pair is not really suitable for Machine Translation, but I believe this is the beginning of a transformation in the way in which we can translate.

I think at SDL with our expertise both in CAT tools, as well as Machine Translation, can continue to enhance the way these two technologies work together.

Want to learn more about SDL AutoSuggest 2.0? Check out Daniel Brockmann’s, SDL Trados Studio Product Management Director, blog here.