Matt Grech
Content Marketing ManagerSmartling
Machine translation is amazing software and the technology has made considerable strides to becoming a more dependable service. Neural based machine translation (NMT) engines, more recently, has transformed the machine translation space and introduced a new level of proficiency.
By attempting to learn the rules and nuances of language, much like our own brains would, these engines can better understand what's being said -- as opposed to simply swapping words based on a dictionary (read: literal translation).
Machine translation should be used, and can be used, for a lot of content. Its so cheap and fast that it wouldn't make sense to just ignore it at this point.
But that doesn't mean its perfect just yet, and when quality is so important, you'll want to set yourself up for success without making any sacrifices.
The major benefits of machine translation
The benefits of machine translation are fairly clear at this point. Input your content and these engines spit out a translated version.
Machine translation is cheap, fast, and quality has dramatically improved in the last 5 years or so.
And NMT is so cheap that you're going to be paying fractions of a cent for each word, with users generally charged for the entire bulk of their translation job. For example, it might cost under $100 or so to translate a couple of thousand words.
There's also that element of a black box in away. You put your content in and out comes a translation. No human intervention, no extra work involved, just a simple and instant process.
Set your Machine Translation expectations
And as we know, machine translation certainly isn't foolproof. Sure, NMT has pushed the technology into a new realm of usability and quality, and improvements continue to be made every day. But it's not foolproof.
Our developers and very own data scientist have spent a lot of time diving deep into different machine translation engines, and the technology behind them, to better understand how and when you should utilize these platforms.
Because different engines work differently, there isn't exactly a one-size-fits-all conclusion to draw here.
Some engines will perform better than others for different tasks or languages. Certain aspects of content can be tricky, and some languages more difficult than others. So, we need to go into developing a machine translation process that leverages those strengths.
Considerations to be aware of when preparing content for Machine Translation
To get the most value out of machine translation, we want to understand where the strengths lie, what they are, and what content will work best so we can work around those limitations for high-quality translations.
MT struggles with overly technical content
Content with a lot of HTML tags and placeholders, a lot of technical aspects, will be difficult to machine translate. These engines can't fully understand what these mean, and it. Doesn't separate tags and HTML from strings
Longer and more complicated strings
Longer, more complex strings that contain more information can be tricky for NMT, which works best when content is easier to understand and digest, broken into shorter, simple strings.
Sentence complexity
As your language gets more complex, MT will struggle with understanding the core message and the surrounding context.
For example, "look at that horse eating grass," may seem simple at first glance in English, but can lead to difficulty in understanding subject-verb agreement for other languages. Is it a grass that eats horses? Or a horse currently eating a fruit?
Specific terminology
Machine translation works best with simple and common language. When you start introducing unique terminology or commonly used synonyms, NMT engines will struggle to determine the proper definition.
Of course, with all of this in mind, that doesn't mean you can't or shouldn't use machine translation. Quite the opposite.
Working around the limitations of MT for high-quality content
If we understand where the strength and weaknesses of NMT lies, we can better prepare our content for translation and only leverage MT when it makes the most sense.
1. Always have a fallback in place
One of the strengths of NMT is simplicity in operation. It’s a completely manual process and happens quickly. So, when things don't go right, you'll want to ensure that your content doesn't come to a screeching halt.
It's critical to implement the right workflows and processes to move content to another translation resource as quickly as possible; preferably a human translator since the content was too complex for an MT engine.
2. Leverage a Machine Translation Post Edit step
Unless this is super low priority content and quality is not a concern (like user-generated content), you'll want to implement a human post-edit step for your content. Without adding much extra time or cost, you'll make sure content is perfect and ready to go without any glaring errors
3. Build Dynamic Workflows to separate content
At the end of the day, some content will require the attention of a human translator. For everything else, NMT is becoming a viable option. Smartling lets you build dynamic workflows to route content to the best resources based on several factors.
You can even configure dynamic workflows to send only some content to NMT, and other strings to human translators.
For example, if you're translating a page of user reviews you'll have the title and header translated by a professional, but all of the user reviews on the page will be handled by MT, and a dynamic workflow an automatically route that content based on tags or labels.
Maximize the value of Machine Translation
Not only is Machine Translation here to stay, but its quickly entering a new realm of usability. We've seen major brands leveraging MT for unique situations like user-generated content, but have also seen brands begin employing MT into their translation stack for quick-turnaround content.
Instead of shying away from new technologies, we want to empower your team to utilize these to their maximum value, and machine translation is one of those solutions that should be a powerful tool inside your toolbox.
If you want to learn more about how you can maximize your translation value, check out our eBook with detailed insight from some of the experts at Smartling!
