What if you could read any newspaper in any language? Or travel abroad without having to worry about misreading signs or menus? Online translation services have been around for years, but they have always been a bit quirky and never quite accurate enough.
New breakthroughs at Google and Baidu are breaking down the language barriers between countries and cultures. In fact, the new technology, called machine learning, doesn’t just make online translation services more accurate, it actually allows the computers to learn and improve.
Why it matters
According to Google, roughly half of websites are in English. But only about 20 percent of internet users are fluent enough to use them. This means that better translation functions will give billions of people access to countless news sources and online content.
But translation has a more immediate use as well. Google has lately noticed a “fivefold increase in translations between Arabic and German,” according to Google chief executive Sundar Pichai. Germany has taken in more than a million refugees in recent years, and being able to use the language is vital to finding work and integrating into society.
So far, Google has installed the machine-learning software for translations between English and Spanish, French, Portuguese, German, Chinese, Japanese, Korean and Turkish. More languages will be rolled out every month.
How it works
Until recently, translation programs took small pieces of the sentence they were trying to translate, looked the words up in a statistically weighted dictionary, and then used a set of rules to figure out where each word should go in the sentence.
Now, with machine learning, the software figures out relationships between words and the algorithm creates a complex web of associations between them. Then, as people interact with the translation program, the computer processes the data and finds patterns in the language that improve its translations.
Machine learning is the future
The heart of machine learning is teaching computers how to find patterns. This means that machine learning won’t work just for translations. Teaching computers to spot patterns will have an almost unlimited number of applications, from improving self-driving cars to recognizing tumors on medical imaging in order to catch early-stage cancer.
For now though, improving translations is a major step forward in bringing the world closer together and helping people connect.