Google has admitted that its translation engines are not perfect and not yet ready for "sensitive debates". Google vice-president Vint Cerf said the use of statistical translation methods, where translations are made on the basis of probabilities and don't rely on parsing, had vastly improved online translation. But he warned about their reliability and said there were problems with interpreting the meaning of the same phrase in British and American English, let alone phrases in different languages.

"I'd be really careful about having any kind of a sensitive debate with someone either spoken or written using these translations," he said during a visit to The Australian's Sydney bureau last week.

Google offers online translations between more than 50 languages. It employs voice recognition in the Google iPhone app. It is seeking to combine these technologies to produce a phone capable of real-time voice translation. Dr Cerf said Google has begun building translation statistics by comparing phrases found in documents produced in multiple languages.

"Now our research teams are starting to combine the statistical methods, which work well when we have a large body of translated materials to work with, to multiple documents written in different languages that purport to be the same thing.

"Our national institutes of standards and technology have rated the Google translations as above all others that I know about.

"If we were going from zero to 10, we would be about five, that's better than almost everybody else.

"But I can tell you that I read newspapers from other countries by using Google Translate and at least I'm getting a pretty good gist of what's being said and if I need to know more I'd go to a language speaker, an expert speaker."

source: http://www.theaustralian.com.au