<< 1 >>
Rating: 
Summary: Still relevant
Review: Machine Translation (MT) was one of the great efforts of computing, ever since the 1950s. The basic idea is that if you have a text string in some human language, and you run an MT program, it will output a correctly translated string in another human language. Notice that this has nothing to do with speech recognition. That is another complicated problem in itself. Well, MT is one of these problems that has proved elusive in solving. Clearly, it should be solvable. Simply by you reading the input, you could translate it into another language that you knew. So we know that a solution does indeed exist. Newton wrote this book in 1992, and it describes MT's capabilities as they existed then. The basic problem of MT seems to be isomorphic to a general AI problem. We (wetware) can translate, because over the years we have built up a world view that lets us do so. Computers have no such thing, thus far. The book describes how practical MT consisted of cleverly restricting the context of the subject matter, so that reasonably accurate translations could be made. Like weather reports, or airline reservations. The book is still useful today. The increase in computing power since 1992 means that its techniques could be applied on far cheaper hardware. Or, that they could be applied with more power, in a brute force manner. But there seems to have been very little advance in the way of fundamental understanding in MT. So the book's methods are still pretty much state of the art.
<< 1 >>
| 
