MOUNTAIN VIEW, Calif.—You might use Google Translate to read a hard-to-find Manga comic book or to decipher an obscure recipe for authentic Polish blintzes. Or, like Phillip and Niki Smith in rural Mississippi, you could use it to rescue a Chinese orphan and fall in love at the same time.
Google is now doing a record billion translations on any given day, as much text as you’d find in 1 million books for everything from understanding school lunch menus to gathering national security intelligence. It translates in 65 languages, from Afrikaans to Yiddish, and can be used on websites, with speech recognition and as an app on mobile phones even if there is no connection.
While the technology is exponentially evolving, Google’s translation guru Franz Och’s face lit up when he heard that the Smiths and their new daughter, 14-year-old Guan Ya, are settling into their new lives together this month communicating almost exclusively through Google Translate.
“All day long I look at algorithms, algorithms and algorithms,” he said. “It is so rewarding to hear that it is touching lives.”
In the Smiths’ case, it changed theirs forever.
The Smiths, who already have three children, first spotted Guan Ya less than a year ago when Niki Smith was looking at photos of hard to place orphans online, offering simple prayers for them one by one. With three children of her own, including a 3-year-old daughter adopted from China, she had no intention of adding to her family.
Then she saw Guan Ya.
“She was just our daughter,” said Smith of that chance Internet encounter nearly a year ago. “There was no doubt about it, from the first time we saw her on the Internet.”
There were seemingly impossible obstacles to adopting the girl. Firstly, Guan Ya was months away from turning 14, the age at which Chinese law would make her ineligible for adoption. Not only could Guan Ya not speak English, she didn’t speak at all.
Guan Ya is deaf.
Undeterred, the Smiths scrambled through the paperwork and home studies that are inherent to international adoptions. With support from both Chinese and U.S. authorities, they expedited the bureaucracy by running a flurry of emails and forms through online translators. And one day Niki Smith received an email from her daughter-to-be, an unintelligible jumble of Chinese characters.
“Well, I couldn’t begin to read this letter,” said Smith.
That is where Google Translate came into play. Smith cut and pasted the letter into the empty rectangle for the program in her Internet browser and Guan Ya’s thoughts magically appeared.
Thus began their heartwarming virtual conversation of love, family and life.
“The computers and software are tools, but I have no doubt that these tools made our bonding so much easier,” said Niki Smith.
Machine translation dates back to the end of World War II, when coders realized that cryptography and deciphering were, in part, math problems. In 1949, influential scientist Warren Weaver laid out a pivotal proposition that paved the way for today’s computational linguistics: a theorem could be developed to solve the logical structure of languages.
Yet almost 65 years since Weaver wrote that “it seems likely that the problem of translation can be attacked successfully,” machine translation is far from perfect.
A team of South African researchers at the Matieland Language Centre recently published a study comparing documents translated between Afrikaans and English by professional translators and then by Google Translate. The results weren’t even close. For the machine-translated writings, “the quality was still below average, and the texts would require extensive post-editing for their function to be met,” they found.
“The general public thinks you can stick anything into machine translation and it’s going to give you everything you need, but of course that’s not the case,” says Jamie Lucero, who heads the translation and interpretation program at Bellevue College in Bellevue, Wash.
He said for high quality translations, literature, marketing materials or complex syntax, a human translator is still essential. But machines are helpful, he said, “for people who just want to get a basic message across.”
And he said some machine translators are better than others.
While translate.google.com leads the market, Microsoft’s www.bing.com/translator offers a similar, free service with 41 languages that users say outperforms Google’s when it comes to language used in high tech and software. WorldLingo is emerging as a leading pay-for-use translator with machine and professional translations in more than 140 languages, for people who require accuracy.
But any translation is a huge leap for communication, said Jennifer Uman , who co-authored a children’s picture book, “Jemmy Button,” with Italian collaborator Valerio Vidali, published this week. They met and then communicated for almost five years on the project almost exclusively through Google Translate.
Initially, she said, the translations were strange. Uman would write, “It looks great,” and Vidali would read “I hoist much illusion.”
“But over the years Google Translate got better and we got better,” she said. “We kind of got the hang of how to use it, and it made it possible for us to collaborate.”
Angolan blogger Rosie Alves, who launched a poetic and often racy blog “Sweet Cliche” a year ago, was confused when she saw that one in four readers were in the U.S. She blogs from her hometown Luanda in her native Portuguese, and counts 18,959 visits to date. In an email exchange with The Associated Press, she used Google Translate to answer questions: “I think it’s very good although the translation is not 100 percent safe,” she wrote. “The best part is knowing that there are people interested in what I write, and use the Google translator to understand my texts.”
Nine thousand miles away at Google’s Mountain View headquarters, Och said he’s hoping to launch translation services for several Indian languages next, including Malayalam, an official language of Kerala state spoken by millions of people. His team consists of computer scientists and programmers, not linguists. And there’s not a single Yiddish-speaking babushka, Basque sheepherder or Vatican-trained Latinist onsite to help.
Instead, the ever-improving algorithms detect patterns in texts already translated by humans, so the more “data” that exists in the form of books and documents, the more accurate the translation. When the system gets precise enough, they roll it out to the public.
There have been a few early releases, said Och. When Iran sank rapidly into an election crisis in 2009, Google released a Persian translator, noting the program was “a work in progress.” And just days after a catastrophic earthquake hit Haiti in 2010, they launched a buggy Haitian Creole translator that, despite glitches, was widely used by rescuers and relief workers.
Last week in a supermarket in Rienzi, Miss., a mother and daughter hovered over an iPhone, passing it back and forth as they strolled the aisles, chatting about what they wanted to cook for dinner. It was just a few short weeks since they’d met in person, since Guan Ya had emailed her family-to-be: “No, I have never been shopping. You do not need to bring me anything. I do not know the things I like. I guess I like chocolate. Have you come to China yet? I will not be afraid. I am very happy.”
When they met, she told her parents that more than anything, she wants to hear. And already doctors in Mississippi are suggesting that hearing aids and possibly a surgically implanted cochlear implant may help.
“So many things had to fall in place on both sides that it is amazing to see how God worked to get her home!” said her mom.
In those early emails, and on this day, the two typed back and forth the three English words, eight simple letters, and the three Chinese characters, a series of strokes and slashes, that mattered the most: “I love you!”