On 1st November, Heike Held and Nikki Elsworth travelled to London for the day to attend this year’s SDL Roadshow. The event held by SDL happens every year and as well as promoting new products, gives a great opportunity for learning more about the industry, the software we use daily and a fantastic opportunity for networking. We managed to meet up with a couple of the translators with whom we regularly work and had a chance to discuss the developments within the industry with translators, project managers and corporates using the software.
This year’s focus was mainly on the industry view of the future, SDL appear to be attempting to incorporate machine translation and post-editing into existing products in order to smooth the transition.
SDL mentioned an ongoing study that they are undertaking, which aims to pinpoint and analyse trends and needs within the translation industry. Some of the key focuses have been the adoption of quality control and how to assess quality, the need for interoperability between software and platforms and the ongoing debate about cloud vs. server based systems.
SDL are releasing information in the format of insights study results, found via their website.
The main updates that we took away from the event was that while machine translation is coming, not very many people are ready to adopt it yet. The risk of storing data in the cloud still seems to be a major concern and this seems to be where most machine translation is coming from. SDL are focusing heavily on the fact that machine translation still requires human input and should be seen as another tool to aid the translator, ensure efficiency and greater quality.
With this in mind, SDL have focused on developing two new aspects to be released with the new version of SDL Studio later on this year (Studio 2017). These are: UpLift technology and Adaptive MT.
The staff at SDL are excited by these two new developments and have aimed to combine cloud and desktop environments using these tools.
UpLift is feature using all relevant resources to aid translator efficiency. It draws on existing data saved in all TM’s, termbases and autosuggest dictionaries that are added to the project. It then uses “fragment recall” to suggest fragments of segments that match the document you are working on. It does this in a “fragment TM” window, highlighting the context from which the fragment has come and also does this in the autosuggest tool, which will suggest the translation after typing the first letter into the target segment.
This UpLift technology searches below segment level and enables the translator to have better TM leverage as it makes use of low matches in the TM that would not usually be picked up as a TM match. It highlights the source match and the target match so that the translator can see exactly where the suggestion comes from and it also learns from the choices made by the translator. i.e. when the segment is confirmed, it automatically updates the TM and the autosuggest dictionary meaning that the information is updated and can be used immediately within the same document and there is no need to constantly export new autosuggest dictionaries. You can manually add terms from the fragment TM window to your glossary as well. Within the UpLift feature, there is also a “fuzzy match repair” feature, which enables certain segments to be translated automatically based on previous sentences and patterns within the sentence, i.e. sentences where the number, colour or product name has changed (but the rest of the sentence has been translated before). Fuzzy match repair should be able to handle changes, deletions, insertions, changes to the order (movement) and punctuation. N.B these segments will be highlighted with a wrench symbol in Studio and will still need to be checked.
AdaptiveMT is a new feature that will be available to translators in version 2017. It is a cloud based self-learning machine translation engine. This works alongside the translation memory to continuously learn and suggest new translations. The difference between translation memory (TM) and machine translation (MT) is that the translation memory is static, it shows and reuses whole segments but cannot separate the words within those segments. Machine translation is dynamic; it learns from patterns within the segments and can take these patterns to create new sentences. AdaptiveMT is designed by SDL to be unique to the user. In the beginning, it will have no data and as you confirm the segments that it suggests, it will learn your preferred style for further segments within the same document and for future translations where you add that engine. NB. You need to check settings, to allow automated translation and select “update” so that the engine will learn. The first engine is free, included with 2017 and further engines will be chargeable. E.g. different engines for different clients will need to be purchased and different language pairs may need to be purchased. The engine is currently saved in the cloud, but SDL assure that the data is encrypted and only stored, not reused.
These two features should aid translators to make full use of their resources and to be able to cope with the tighter deadlines that the industry is seeing as the need for fast data becomes greater.
Aside from this, SDL have also worked on updating GroupShare with new features such as auto save and better support on mobile devices and they have changed the view of MultiTerm to enable better usability and to encourage those who do not work with glossaries, to start agreeing terminology and to be less reluctant to use the tool. As Balthasar already use glossaries and MultiTerm, this change will not affect us too much, other than to make it easier to edit terms within MultiTerm in the future.
The day was very informative and extremely useful. We both came away with new ideas; excited to see what SDL Studio 2017 will be like and encouraged that SDL are continuing to analyse the market and research more efficient ways of working and how to use technology to assist the industry in the future.