Machine translation has been in our collective presence for a number of years now, and depends on who you ask, they are either a tremendous help or a problematic burden. Like any tool in the modern age, its usefulness really depends on how you use it. Machine translations are not accurate, and should not used as a replacement for professional translation services. However, people still use them as such either because they are not aware of how inaccurate they can be, or simply as a cost-saving measure. In either case, you can rest assured problems will always arise.
Although, one company has been working on a machine translator that actually learns languages, in the hopes that in the not too distant future, the machine will be “intelligent” enough to provide accurate translations, on par with that of its human counterpart.
How is this possible, one might ask. Well it all starts with breaking down the process of reasoning (that little element humans have that machines don’t…yet). Once that process is delivered in a way a machine can understand it, it’s just like the process humans go through in order to learn: through experience. Although experience for machines is just a matter of reading.
Tom Mitchell, a professor of computer science and Chair of the Machine Learning Department at Carnegie Mellon University in Pittsburgh, Pennsylvania explains, “Our Never-Ending Language Learner (we call it NELL, for short) is a computer program that is learning, 24 hours a day, to read the web. Each day NELL has two tasks. First, it must collect more factual beliefs by *reading* text it finds on the web — beliefs like “Obama is president of the US,” or “T-shirts are often worn with blue jeans.” Second, each day NELL must *learn* to read better than it could the day before, so that it can extract more facts, more accurately tomorrow.”
NELL has been running nonstop like this for over three years, and the result so far is a collection of 70 million beliefs NELL has read, which it holds with different confidences. And, NELL is learning to read better – it is a better reader today than last month, and much better than last year.
Beyond learning to read, NELL is also now learning to *reason* — to draw its own conclusions from the facts it has read. How? By analyzing the 70 million beliefs it has read, NELL can discover regularities that enable it to infer new beliefs. For example, NELL discovered that usually “IF a person plays on some team, AND that team plays baseball, THEN that person also plays the sport baseball.” NELL uses this self-discovered common-sense rule – and tens of thousands of others – to reason its way to new beliefs that it has not yet read.
You can follow NELL’s progress, and give it some guidance as it tries to learn, by visiting its website.