In 2012, Microsoft’s Chief Research Officer Rick Rashid, did a live demo of an instant translation of his speech from English to Chinese. Consequently, Skype (owned by Microsoft) released a preview in December 2014 of Skype Translator, an app that people can use while making calls to translate a conversation in real-time. Currently, it only translates for people exchanging in English and Spanish. Initially, users have reported that the translation isn’t entirely smooth or quick. However, it will be a matter of time before the kinks are worked out and Skype will add more languages into the service.
It seems that a universal translation tool touted in sci-fi is within our reach. How are researchers able to achieve this important breakthrough? They have been utilizing a field of machine learning called neural networks, where a computer recognizes patterns from large and ambiguous sets of data.
One way to see how this is applied is through speech recognition. How does a machine recognize when someone says, “hello?” Well, they take audio samples of thousands of different people saying “hello.” The computer notices a pattern in the way the sounds are produced to predict and recognize the next person that says hello.
On top of the layer of sound, the machine needs to have a robust index of words and phrases. However, translation isn’t exact and also requires an understanding of context. It is useful for a machine to understand the sentiment the user is trying to convey as well.
Google is trying to apply neural networking techniques to solve these problem using Word2Vec. This program assigns “vectors to words, which show a word’s relationship within its own language.” Instead of assigning a 1:1 relationship between a word and its meaning, it could look at how a word is used in conjunction with other words (hundreds or thousands of them) to change its connotation. It can be used to translate even “unknown words by comparing its vectors to vectors of known words in other languages.” What is interesting about neural networks is that the more languages that are added to the software, the better the translation service gets.
For many years, I have enjoyed a slight edge for being bilingual. It certainly has helped me in my business transactions in China. However, my husband started a business a few years ago and had buyers from around the world, wanting his product. He overcame the language barrier by using Google Translate for his email communications. I wonder how many more sales he could have made if he could directly converse with his customers using an instant audio translation app?
Just this week, my mom has been bugging me to teach my children Chinese. She thinks I should enroll them in weekend Chinese School classes. I wonder if it’s even necessary (Heresy, I know!). The technology is so close, and I predict that within 5 years, the software will provide perfect and instant translation in any language they choose.
I believe that even if machines can provide contextual meaning in translation, they probably will not be able to convey cultural norms, body language, tone, and etiquette—at least in the near future. All of those cues for language take practice and probably are best experienced in person.
What happens to language studies when the technocrats in Silicon Valley crack the language nut? Does this mean we no longer need to study Spanish, Chinese, or French in school anymore? Will language studies go away like cursive handwriting?