No matter what area of business you’re in today, it’s hard to keep up with the accelerating pace of technological change. The onslaught of new devices, new apps, and new tech buzzwords never seems to end. Translation services are certainly no exception. If you pay even a modest amount of attention to the news, you’ve certainly heard about advances in machine translation, cloud-based translation platforms, proxy services and specialized QA tools. Technology evangelists tout their myriad benefits, while often failing to caution adopters on potential costs and pitfalls.
At MTM LinguaSoft, we understand that despite the increasing pace of technological innovation, translation is and will remain a human-centered process.
If the unique selling proposition of a language service partner (LSP) is their exclusive use of cutting-edge technologies, potential clients should beware that a strong focus on new technology may be happening at the expense of human talent.
Experienced, top-notch translators may not want to spend time learning to use the latest “hot” translation tool, because if a particular LSP requires it, they can find work elsewhere.
We take a wide-ranging approach to translation technology, maintaining expertise in a variety of tools so that we can assemble the right translation team for a project, and then choose the combination of tools best-suited for that team and that project. Here’s a summary of our current take on translation technologies that boost quality and efficiency without getting in the way.
Cloud-based translation platforms
Cloud-based translation tools store the base application and all working files on remote servers accessed via the Internet rather than on the user’s personal computer. This carries a number of real and potential benefits for service providers and clients alike.
Benefits of cloud-based translation
The main benefit of cloud-based translation platforms is their collaborative workflows, which makes it possible to dramatically shrink turnaround times while improving quality and efficiency.
- Translators have access to the file and their language tools wherever they are, making it easier for them to meet deadlines.
- Multiple translators and editors can work on a translation simultaneously and share their collective knowledge and cognitive resources. No time is wasted sending a file from one resource to another.
- There is a “two heads are better than one” effect even with basic two step projects. With more pairs of eyes on a project at the beginning, key questions get asked and resolved more quickly.
Drawbacks of cloud-based translation
For translators, potential downsides of cloud translation include possible loss of control over translation memories and potential increases in costs. License fees for Memsource and Memoq cloud, for example, can be quite expensive. Fortunately, the rapid pace of translation tech innovation includes new business models that mitigate these potential downsides for translators.
SmartCAT, a new cloud-based solution developed by the same company that produced the popular ABBY Finereader (though they have since split), offers state-of-the-art translation technology at no cost to the translator. How do they make money? Their business model encourages users to use their full-featured invoicing and payment system, for which SmartCAT takes a 5% cut. They also offer embedded services like Lilt’s machine translation and industry-leading OCR functionality, for a small fee.
Translators hoping that machine translation (MT) would simply go away if they chose to ignore it are out of luck. With each passing year, progress in machine translation seems to take a giant leap. But there’s no need to worry. Despite rampant fear-mongering, machine translation is more likely to expand the role for human translators than put them out of work. Why?
According to a 2016 Common Sense Advisory Study, the percentage of new global content that gets translated each day is vanishingly small, less than a millionth of a percent. As machine translation improves, it drives down the cost of translation and increases the likelihood that businesses will view translation as cost-effective. And while the quality of machine translation continues to improve, it won’t be able to replace human translators in business-critical applications for quite some time. And as the use of machine translation expands, the need for human post-editing will grow right along with it.
Three phases of Machine Translation development
MT has moved through three major technological phases, beginning with rule-based translation, then statistical machine translation and finally the current and much-hyped neural machine translation.
Rule-based translation (RBMT) attempts to mimic the logic and syntax of source and target languages and make word-for-word substitutions using dictionaries. While the concept behind RBT may be the easiest to understand from a human point of view, results were disappointing. Translations were often wooden and required a great deal of post-editing to be usable.
Statistical machine translation (SMT) generates translations based on their statistical likelihood, using a large corpus of paired translations as data. SMT generates superior translations to RBMT, but it does not do well with specialized domains or words that are used infrequently. Its reliance on existing matched pairs can sometimes generate weird results.
Neural Machine Translation (NMT), currently all the rage, attempts to model the learning structure of the human brain, with digital neurons whose connections get stronger or weaker based on feedback. It can generate translations that appear very natural, especially within very general domains and with major European languages, but it exhibits weakness similar to SMT with specialized domains and uncommon turns of phrase. And while NMT translations sound more fluent than their SMT counterparts, SMT translations still appear to have the edge in accuracy.
This last point about NMT is worth keeping in mind if you are taking a job post-editing a machine translation. While NMT post-editing may be easier overall, one has to be careful about seemingly fluent translations hiding critical inaccuracies. No matter how accurate the machine translated output appears, the editor will need to compare carefully with the source text.
Until recently, the technical expertise and data resources necessary to pull off usable machine translation was prohibitive for freelance translators. That’s starting to change with some service offerings Lilt and MateCAT. And the up-and-coming SmartCAT cloud service offers Microsoft NMT in addition to Lilt integration.
Translation Quality Assurance (QA)
QA tools are algorithms that can identify possible errors in translation. The algorithms do not (usually) include any “understanding” of the target language; instead, they look for patterns in the source and target segments that suggest error. Standard CAT tools like SDL Studio and MemoQ contain built-in QA tools, but they have their drawbacks. Translators often complain that MemoQ QA returns too many “false hits” – flagged errors that are not errors – while SDL Studio often has the opposite problem.
For better QA, translators are increasingly turning to third party tools, whose sensitivity to error can be tweaked and balanced for better efficiency and utility. Among the market leaders in QA tools are QA Distiller, Verfika and Xbench. We find Xbench to provide the best balance of cost and utility. In addition to providing powerful search functions and custom checks, XBench allows you to publish a report showing QA issues that were identified and how they were resolved. Delivering a finished translation along with a QA report shows that you have “gone the extra mile” to deliver an error-free translation.
Staying Ahead of the Curve
Translators can benefit greatly from wise use of new technologies. How does one know, though, whether the latest tool or application is a critical technology or a needless distraction? Does the new tool require a steep learning curve or is it easy to use right out of the box? One should be wary of software with short history and big promises. Before investing time or money in new technology, have an idea how it will impact your workflow and your productivity. Have a plan and a timeline in mind, and don’t be afraid to admit failure. And remember, translation and communication will remain human centered processes, at least until Skynet becomes sentient and the AI robots rise up and take over the world.