A slow translation process can be a real drag on your global plans. It holds you back, creating unnecessary obstacles on the road to international growth.
Because of this, more and more companies are looking to machine translation solutions to speed up their translation turnaround times and get more content to market much quicker.
Machine translation has come a long way in terms of the quality levels it can produce, too (which is another reason why its adoption rate continues to skyrocket). This is, in part, due to the work that’s done under the hood: custom-training the engines.
How are these engines trained to run like the well-oiled machines that they are? If you’re involved in this training process, you might wonder how these engines continue to produce better quality translations over time.
Ladies and gentlemen, start your engines. We’re going for an information-packed ride.
Parts needed for training machine translation engines
First off, the base of a custom engine is a foundation engine. Experienced translation vendors have access to foundation engines which help build the machine you wish to use in your translation process. These foundation engines focus solely on one language pair and one vertical, such as German-to-English engineering or English-to-Japanese marketing.
By pairing these foundation engines with your company’s previously translated content, a machine is more properly trained to translate your content in particular—learning your brand terminology, style preferences and more.
To be the most successful at the starting line, it helps to have a large amount of high-quality, previously translated content (which can be your translation memory).
Other multilingual resources that are extremely helpful include:
Do-not-translate lists
If you don’t have these materials or a lot of translated content to work with, don’t fret. You can still train a machine translation engine; just know that it will take little longer for the engines to learn your company’s brand voice.
Typically, this training process, in which we run your content through the engine and test (and re-test) the initial output, takes about four to six weeks.
Regular maintenance allows engines to continually improve
The training process doesn’t stop there, though. Your translation vendor will continually monitor your engines over time to improve the quality of the output even more. After all, everyone makes mistakes—even machines.
The beauty of working with a machine, however, is that its mistakes are consistent. So once they’re fixed, your translation provider can resolve the error forever (meaning that pesky word choice won’t show up your translation output anymore).
How are these errors corrected? If you have a post-edit step after machine translation, the linguist’s corrections automatically update into your translation memory. Therefore, whenever your translation memory is applied, the correct version of that segment will end up in your translated content.
Think of machine translation engine training as a process. In other words, clients that find machine translation to be successful consider early results to be indicators, not final examples of quality. The more work that is run through your machine over time, the better it becomes at driving accurate translations for you.
A general rule of the road
Some raw machine translation output may be of acceptable quality without any editing. This really depends on your unique needs, definition of quality and the engine maturity level.
It may not be sufficient for public-facing content though. At Sajan, we always recommend some kind of post-edit support for machine translated content that will be used externally. This extra quality measure safeguards your brand image so you can avoid any unexpected run-ins with uneven brand identity across markets.
Want to learn more?
This is just a quick look under the hood into the machine translation training process, but it should help you understand how advanced machine translation differs from free instant translation tools on the internet.
If you care to learn more about machine translation best practices and how the technology works in the corporate communication setting, leave your questions below.
Val is on the Advisory Board for the Technical Communications Program at the University of North Texas. When not working with customers or students, Val can be found sitting behind her sewing machine working on her latest quilt. She also makes a mean hummus.
- How to Make Conferences More Inclusive for the Hard of Hearing Community - December 2, 2024
- Preparing Content for AI: 6 Reasons Why You’re Not Ready - August 29, 2024
- How to Be Inclusive in the Workplace: My Experience as a Hard of Hearing Person - August 12, 2024
