Machine Translation (MT) has evolved to be more accurate, more nuanced, and more reliable than in years past.
Linn Hu and Chris Wyant explain how content managers can incorporate MT into their translation process. In addition, they cover what types of content to translate with MT, along with how to choose the right MT solution for your content.
The Evolution of Machine Translation
MT has evolved significantly from traditional phrase-based MT - grouping words into phrases and then translating by recognizable phrases - to neural MT. Neural MT is the current standard of MT, which looks at the context of a full sentence to translate words not as individual pieces or chunks, but as pieces of a larger puzzle. Thanks to neural MT, machine translations have become much more accurate, with one-off errors here and there rather than large chunks of translations that do not make sense.
When to Use Machine Translation
When thinking about how to choose content to machine translate, it's important to consider the 1) geographical value, 2) product/content value, and 3) the customer journey stages you want to address. Ideally, MT works best if you're testing the waters of a new market, and don't yet want to invest heavily in professional human translations. In addition, if you have content that brings in significantly lower revenue than other content, this would be a good contender for MT. Lastly, MT works well for straightforward support content but not for highly creative brand and marketing content.
How to Evaluate MT Quality
How do you evaluate MT quality between providers? At Smartling, we incorporate MT Post-Edit - which entails having a human editor review machine translated content and edit as necessary - and track how much the editor had to edit/change the content to reach a final, acceptable output. The less editing required, the better the MT quality is.
For appropriate content types, machine translation can be a highly cost-effective way to create new translations at scale. Many Smartling customers have begun incorporating it into their localization program, and we expect to see more to come.