Breaking Language Barriers

                                                   
Your voice in language 1 – MachineTranslation - Your voice in language 2

Last May I blogged about the ways in which natural language processing is changing our world, mentioning a number of applications, including automatic machine translation of speech using one’s own voice. I wrote that computer speech synthesis had advanced to an extent that, in the near future, systems would be able to translate your speech using your own voice. Theoretically speaking, you would be able to hear your voice speaking a foreign language without your necessarily having learnt that language. With sufficient samples of your speech, such systems would be capable of putting together new sentences for you, in the new language. The systems would just need to know your voice, and they would do the rest of the work.

Well, that future is here now. Rick Rashid, Microsoft’s Chief Research Officer, has recently demonstrated automatic translation of his English speech into Chinese at Microsoft Research Asia’s 21st Century Computing event. One of his blog posts includes a recent video of his demonstration, entitled “Speech Recognition Breakthrough for the Spoken, Translated Word”. In that post, he explains that the first challenge in such systems is for the computer to actually understand what you are saying – a challenge acknowledged by experts decades ago; the past decade has seen breakthroughs reflected in a “combination of better methods, faster computers and the ability to process dramatically more data” (“Microsoft Research Shows a Promising New Breakthrough in Speech Translation Technology”).

Rashid adds that over two years ago Microsoft researchers made their systems far more intelligent by using a technique called Deep Neural Networks that simulates the way the human brain works. He asserts that error rates in speech recognition have been reduced from around one in five words to about one in eight, and that, similarly, machine translation has become more reliable. In the case of English to Chinese, the system works in two steps: first, it translates the English words into Chinese equivalents; then, it re-arranges the words to form proper Chinese sentences. While acknowledging the persistence of errors, including amusing ones, in both the English text and the translation, he projects that, in a few more years, these systems will be astonishingly reliable.

This is definitely breaking news on breaking language barriers. The implications for academic institutions might warrant some consideration.