Language translation is a notoriously difficult task for humans , let alone electronic computer . But in trying to work out that job Google has stumbled across a canny magic , that involves plow them like mapping — and it really , really works .
Technology Review has shone a lighton a Google research paper release on the arXiv host . It spells out how the hunting giant ’s most late translation study focus on plotting out discussion on language maps — recall of a space full of parole , rate in some sort of logical order — and simply uses linear operations to shift between language . Technology Reviewexplains :
The new approach shot is relatively straight . It relies on the notion that every language must identify a interchangeable set of ideas , so the words that do this must also be exchangeable . For example , most linguistic communication will have words for common fauna such as cat , andiron , cow and so on . And these intelligence are in all likelihood used in the same way in sentences such as “ a cat is an fauna that is small than a dog . ”
The same is dependable of numbers . The image above usher the vector representations of the numbers one to five in English and Spanish and march how exchangeable they are .
This is an significant clue . The novel trick is to represent an total language using the family relationship between its words . The set of all the relationships , the so - called “ language space ” , can be remember of as a set of vector that each point from one word to another . And in recent years , linguists have discovered that it is potential to handle these vectors mathematically . For model , the operation ‘ Rex ’ – ‘ man ’ + ‘ fair sex ’ results in a vector that is similar to ‘ queen ’ .
turn out that muckle of languages share a huge identification number of similarities when they ’re mapped out in this way , which means the problem is n’t about finding the good words , but about detect the good mode to represent one vector space to another . In other words , version ’s no longer a problem of linguistics , but of mathematics .
The neat thing is that the technique does n’t really make any assumptions about the languages involved , it just interrogates the fashion transmitter blank are related to teach other . That makes it fabulously versatile . But the algorithms are still in their early days ; there ’s a way to go yet . [ arXivviaTechnology Review ]
figure bydylancantwellunder Creative Commons license
GooglemathScience
Daily Newsletter
Get the ripe technical school , science , and finish news in your inbox daily .
News from the time to come , deliver to your present .
You May Also Like
