SDL has put translator productivity at the heart of its latest major release, the 2017 SDL Trados Studio and GroupShare update. The Studio 2017 release represents significant innovation in an area of the translation supply chain that, perhaps, has not received its fair share of attention over the past few years.
Translation Memories contain the result of tens of thousands of hours of intense knowledge work. So, the challenge SDL’s engineers gave themselves is how to make this treasure trove of data and expertise more readily accessible to those who work with it on a daily basis: the linguists. The result is upLIFT, a powerful concordance search that unlocks productivity gains from no-matches and fuzzy-matches in new, exciting ways.
Another major challenge SDL set out to address was the historical reluctance by many translators for interacting with machine translation output. One of the biggest pain points has always been that the machine never learns and the time spent post-editing never shortens. Enter AdaptiveMT, a machine translation solution personalized for each translator that learns from past mistakes.
Both upLIFT and AdaptiveMT are integrated into the SDL Trados Studio 2017 and GroupShare release. SDL’s terminology management suite, MultiTerm, also received a major facelift and is more user-friendly than ever.
Meet upLIFT: Where Concordance Meets Terminology
Accelerating translation productivity is all about maintaining the flow. Each time the translator has to leave the existing user interface and go look elsewhere for information, there is a small break in focus and the mind drifts.
upLIFT leverages the treasure trove that is the Translation Memory to increase translation speed for no-matches and fuzzy-matches. The “LIFT” in upLIFT stands for “Leveraging Intelligent Fragments from Translation Memory” to reflect this new benefit.
With upLIFT, SDL takes the concept of phrase-level matching that worked well in AutoSuggest and turbocharges it. AutoSuggest requires TMs of a certain size to build a dictionary, while upLIFT does not.
In upLIFT, it is now clear which target phrase corresponds to which source phrase. And while a concordance or TermDB search has the translator looking up from the segment, upLIFT displays smoothly as you type. Furthermore, upLIFT does away with the extraction step. Translation Units are available for subsegment recall as soon as they enter the Translation Memory. From now on, think of your Translation Memory as your lightweight termbase.
Finally, in a big step forward for translators in some of the world’s most dynamic markets, the 2017 rollout includes upLIFT for Chinese, Japanese, and Korean.
Beyond phrase recall, SDL looked into how fuzzy-matches could be made more relevant to the translator. After all, if the system can help a translator improve the quality of a fuzzy-match automatically, it should. And so upLIFT Fuzzy Repair was added to Studio 2017. Fuzzy-matches are no longer static but can now be repaired by tapping recalled phrase fragments or other sources, such as termbases and machine translation. This means less typing for the translator and improved productivity.
AdaptiveMT: Learning From Linguists on the Fly
Machine Translation has made impressive progress in recent years and is set for another quality leap with the advent of neural machine translation. Yet, despite the headlines, the future belongs to complementing, rather than replacing, human translators with the latest machine translation technology.
A common accusation leveled at post-editing has been that the machine translation output does not incorporate changes made by a post-editor. Over time, this has led to frustration among translators and put the brakes on human-machine interaction in translation.
It’s all about maintaining the flow
With the introduction of AdaptiveMT, the paradigm has now firmly shifted. The machine translation engine no longer ignores corrections made by linguists. It learns on the fly. SDL’s AdaptiveMT notes every correction, stylistic preference, and terminological choice so as to improve output over time. SDL secure AdaptiveMT adapts to each and every translator. It also does away with the need for manual training of a machine translation engine.
The live, learn-as-you-go private personalization of a machine translation engine by an individual translator is a radical change for the better. From a user interface perspective, SDL AdaptiveMT means translation memory and machine translation are converging for the benefit of human translation productivity.
The Dawn of a New Reporting
The positive impact of SDL’s upLIFT and AdaptiveMT goes beyond the translator’s desk to areas as diverse as sales or accounting. With the two new features, SDL has made a major update on its file analysis report, which often forms the basis of how translators and language service providers price and invoice their work.
In addition to familiar categories such as perfect match, repetitions, 100% matches, and fuzzy-matches, the new report leverages upLIFT to provide detailed information on fragments recalled from either whole translation units or fragments thereof.
Based on AdaptiveMT, the new report now contains two new categories that have the potential to change the way the industry thinks about pricing. The “MT baseline” category captures content, which was translated using the baseline machine translation engine and does not yet contain personalized input by the human translator. The “MT With Learnings” category, meanwhile, denotes content for which the machine translation engine has leveraged previous human input to improve quality.
Using a so-called “TER” (Translation Error Rate) algorithm, SDL predicts how much less effort is required for post-editing a segment that contains self-learned phrases compared to a baseline segment. SDL recognizes that predictions on gauging productivity gains represent a work in progress, and that the potential impact on industry pricing will be significant.
Innovation at the Heart of the Supply Chain
Over the course of the coming months, SDL is rolling out these exciting productivity boosters across a range of language combinations and products. In the first half of 2017, SDL GroupShare 2017 will add support for upLIFT technology. SDL is also working on combining these new technologies into one single cloud-based service called “Best Match”. Best Match combines AdaptiveMT with cloud based translation memories and terminology repositories to calculate the best possible match that requires the least editing effort.
SDL delivers innovation at the heart of the translation supply chain, where the linguists do their work.