Existing statistical translation-based online tools such as Bobby Fish and Google Translate, translate the results barely available, but still far from perfect. Now, Google engineers preprint papers published on the website, presents a statistical translation does not rely on new technology.Their method is to translate into a vector space language mathematical problems, the use of data mining techniques modeling the structure of a language, and then with another language structure comparison.
The new method is based on a concept that every language has a similar concept, which uses the word the way there are also similarities. For example, are described in most languages the same animal words, such as dogs, cats and cows, these words are likely in the same manner used in a sentence, such as “cats are smaller animals than dogs.”The relationship between words in the language set is called “language space” can be assimilated to a point to another set of vectors, can be mathematically processed, for example, male plus female equals minus king queen (‘king ‘-‘ man ‘+’ woman ‘=’ queen ‘).In the vector space, many languages have in common, we can use vector space linear transformation translate one language into another language. Google engineers said that the new method is very effective.
This article is reproduced from: Solidot
(site only for reprint site does not mean that the text agreed views or confirm the information contained herein)