Using a computer-assisted translation tool, otherwise known as CAT tool, is a process through which individuals utilize software in translation to save time and money, and possibly better the quality. Among numerous CAT tools available in the expanding industry, including memoQ, Across, and Wordfast, I particularly focused on SDL Trados to educate myself on use of CAT tools and officially enter the world of machine translation. The goal of translation technology is simple: achieve fully-automated, unrestricted, high-quality translation product. Of course, that would ultimately mean the domination of artificial intelligence in the industry. And considering the recent improvements of Google Translate, that future seems to be well under way. Regardless, I hereby present a specific CAT tool project I worked on with a group of Korean translation, interpretation, and localization management students to simulate a real-world translation project. Following the briefing of our group’s project on the game Sherlock Holmes: The Devil’s Daughter, allow me to gloss over my personal observations and predictions of the upcoming technology in the future of translation industry.
STATEMENT OF WORK
Description of the Project Specifications/Goal
- The project requires translation of the manual of the game Sherlock Holmes: The Devil’s Daughter in order to raise the Korean audience’s interest of this particular game. Another purpose is to introduce the premise (history, plot, etc.) of the game as well as the characters to the Korean audience. This project will illustrate the uniqueness of the Sherlock Holmes world, and elaborate on the specific qualities of Sherlock Holmes game play that distinguishes it from all of its competitors.
- 5 Translators
- Computer-Assisted Translation Tools: SDL Trados, SDL MultiTerm
- Statement of Work (SOW) Release: Nov 11, 2016
- Client Kick-Off Meeting: Nov 16, 2016
- Project Start Date: Nov 16, 2016
- Translations Due: Nov 17, 2016
- Compiled Translation Draft Due: Nov 18, 2016
- Final Delivery: Nov 18, 2016
Outline of Preparation/Setup Phases and Steps
- Each translator receives individual source text files from the engineers. Engineers tend to inform translators which translation tool to use based on a client’s request.
- Check if there is any protocol or style guides to use as a reference. If no glossary, protocol, or term base is given, translators perform term extractions and research.
- Translators create a Trados project and TM accordingly and make sure TMs are used universally among all translators for consistency.
- Translators translate the source text while performing a research and adding terms to the term base.
Outline of Finalization Phases and Steps
- Once translators finish, use the Quality Assurance settings in Trados to check for errors and ensure quality. All translators should use the same setting adjustments for consistency.
- Export reviewed target text from Trados as a Word document.
- Save TM in both the Trados-specific format (.sdltm) and universal formal (.tmx).
- Export glossary termbase as an Excel spreasheet using MultiTerm.
OVERVIEW OF LESSONS LEARNED
The purpose of this project was 1) to select a topic that would be applicable to the real-world, 2) capture the heart of the Korean audience, and 3) help them immerse into the world of Sherlock. Instead of choosing a topic that we’ve already worked on in other translation classes, we decided to research and choose a topic that may be applicable to a real-life situation.
With the recent trends of the Korean entertainment gearing towards mystery, crime, and thriller content, Sherlock was undoubtedly one of the best choices. And as Korea has one of the biggest gaming industries in the world, so we can easily say that we took full advantage of this opportunity to simulate the whole process as if we are working on a real translation project.
- Added terms on MultiTerm.
- Created a new project.
- Translated the source text.
- Updated Translation Memory.
- Edited the translated file (“Open for Edit”).
- Saved (exported) the translated file as a word document (.docx).
- Worked on alignment.
- Exported files (terms, translation memory, etc.) in different formats (Excel, TMX, .sdltm).
ROOMS FOR IMPROVEMENT
Because we were working with multiple translators, it was important for us to have a consistent termbase and translation memory. However, because Trados was lacking in this particular “share” function, we had difficulty figuring out how to save time and use the same termbase and TM.
Also during the client meeting, we were asked when a third party editor can get involved in the project. We naturally thought that it would be after we submit the deliverables to the client, meaning after all the translation was complete. However, the client — a.k.a. our professor — advised us to get the editor involved as soon as possible. While it may be good to get the editor involved as soon as possible, we thought that this may bring confusion, especially in a project with multiple translators. It is good and definitely necessary to always keep the editing process in mind, but the involvement of the editor should be well-time so as to cause as little confusion as possible.
We also had errors with hyperlinks/URLs that prevented us from a successful alignment. From the source text, we had about three hyperlinks to a gameplay demonstration video, but when we submitted the translation into Trados, the system failed to recognize the URLs. So when you take a look at the screenshot below, notice that “A Master of Disguise” is on line 8. But on the Korean translation side, “A Master of Disguise” is on line 7.
We later attempted to solve the issue by manually aligning each sentence (which was, as you can imagine, extremely time-consuming), but because Trados failed to recognize or read the link at all, alignment still did not work.
Other than simulating a potential real-world project, we successfully managed to work with MultiTerm as well. Prior to translation, because we had already entered specific phrases and/or vocabularies into MultiTerm, the system was able to provide such information as soon as it recognized we were translating the complementary terms. This way, we were able to save time but still maintain the high quality of our translation.
It would be much easier for the group of translators who work on the same project to use SDL Studio Groupshare software. This would let translators work on a single project simultaneously and even let them use and update the TM.
NOW WHAT? THE FUTURE OF TRANSLATION TECHNOLOGY
Artificial intelligence is coming. In fact, it’s coming fast. With technologies like machine learning, deep learning, and big data, artificial intelligence has been evolving at an impressive speed to an impressive quality. Earlier this year, Google presented AlphaGo to play the game of “Go” against the world champion, Lee Sedol. To everyone’s surprise, AlphaGo defeated Lee Sedol with flying colors, resulting in 4:1 in a match of five rounds. It was a pivotal moment that garnered the world’s attention towards the unimaginable potential of artificial intelligence.
So how will artificial intelligence and the upcoming technology affect the translation and interpretation industry? As Ray Clifford, Associate Dean of Humanities and Director of the Center for Language Studies at BYU, once said, “Computers will never replace translators, but translators that use computers will replace translators that don’t.” Of course, now with the recent developments in Google Translate (Korean into English translation is quite great), there is a chance that computers may replace translators and produce editors instead, through processes like post-editing. Depending on the genre of text, however, artificial intelligence will not be replacing translators any time soon, primarily in literature and other works of art.
Although I cannot presume how quickly AI will be able to attain a truly human-like mind, I doubt that it will perfectly embody the complicated levels of human emotions that even humans don’t seem to fully understand. Frankly speaking, I believe it is simply the purity of human emotion that prevents AI from dominating translation, interpretation, or any other types of human communication. Machines may be able to “self-teach” themselves through machine learning, deep learning, and big data, but ultimately, humans are the ones who effectively insert the necessary information into artificial intelligence. Hence, I see a pool of what experts would probably like to call “premium” translators and interpreters surviving in the long run, to not only produce quality material but also to educate artificial intelligence.
The evolution of AI is not a matter of technology taking over humanity, but a matter of coexistence between humans and machines. We just need to find a way to support the existence and purpose of one another.