Building an Unlocalizable Product
This is the problem I see most often. For the most part, localization involves the optimal use of language. Messages, labels, calls to action—they are all text and need to be translated. But when you initially build a product from the ground up, it’s all too easy to forget about localization, hardwiring all the text into markup or code. Unless you make sure to factor out these assets from the beginning, it will be expensive even to discover them when you need to add the first localized language.
A particularly notorious offender is code that glues together messages programmatically—for example, “You have [ x ] items in your [ y ].” There is virtually no language on the planet other than English in which it’s possible to just translate the strings You have and items in your and expect the result to make sense.
Unless you engineer your product to be localizable, all your efforts are doomed to fail. Making an early investment in creating a localizable product will pay for itself several times over down the road.
Using Machine Translation
Machine translation (MT) has profited immensely from the ongoing deep-learning revolution. We have seen an unprecedented breakthrough in quality over the last two years alone.
While MT has crossed a threshold and now enables humans to grasp the true meaning of text in a different language, it is essential to remember that MT does not possess human intelligence. It lacks understanding, empathy, and nuance. Its output is always ungrammatical, imprecise, or at the very least, unnatural. Just plugging machine-translated copy into a product is guaranteed to alienate the people engaging with the product. Using machine-translated text inevitably makes products cryptic or even downright unusable.
If you are tempted to use MT to save costs, remember the impact replacing Submit with Buy Now has had on conversion rates. MT misses all such nuances and a lot more.
Providing No Context for Translators
One reason today’s MT systems cannot hope to achieve human quality is their lack of context. There’s no way to tell an MT system that a particular instance of check out is a label on a button that initiates a purchase rather than a sign on a hotel-room door.
Even when you’ve made the right choice and hired a professional translator, they cannot deliver the desired outcome if they lack context. Throwing an Excel sheet at a translator, with row after row of isolated user-interface strings is a sure way to get inadequate translations.
Go after the lowest-hanging fruit by providing meaningful identifiers that help place every piece of text within the product’s context. For the best outcomes, invest in technology that allows translators to view an actual app or Web site to see every label, message, and ToolTip as it appears to users in context.
